Patent Application
20220246170
This disclosure relates to the technical field of audios, and more particularly to a method and apparatus for detecting a valid voice signal and a non-transitory computer readable storage medium.
Voice is as a means of human-computer interaction. However, noise interference always exists in a working environment, and the noise may affect application effect of voice. Therefore it is necessary to detect a valid voice signal and distinguish the valid voice signal from a noise interference signal for further processing.
A difference between the voice signal and the noise signal can be reflected in energy. In the case of a high signal-to-noise ratio (SNR), which can be regarded as a ratio of the voice signal to the noise signal, the energy of the voice signal is generally much higher than that of the noise signal. However, in the case of a low SNR, if noise frequently appears in an input audio segment, the energy of the noise signal is relatively high and is almost the same as that of the voice signal. In the related art, a method for detecting a voice signal based on signal energy is adopted to distinguish the voice signal from the noise signal according to short-term energy of the input signal. That is, energy of an input signal in a time period is calculated and then is compared with energy of an input signal in an adjacent time period, to determine whether the signal in the present time period is the voice signal or the noise signal. According to the scheme of the related art, calculate and compare energy of signals in time periods. Due to frequent appearance of noise, noise appears in the signal in the present time period and also appears in the signal in the adjacent time period, energy in the present time period is a sum of energy of the noise signal and energy of the voice signal, and energy in the adjacent time period is also a sum of energy of the noise signal and energy of the voice signal, so existence of noise cannot be detected through comparison. The frequent appearance of the noise makes the energy of the signal increase, which may affect detection of the signal, and accordingly it may be possible to regard the noise as the valid voice signal, so in the related art detection of the valid voice signal may not be accurate.
In view of above problems, a method, device, and apparatus for detecting a valid voice signal and a non-transitory computer readable storage medium are provided in the disclosure.
According to a first aspect, a method for detecting a valid voice signal is provided in the disclosure. The method includes the following.
A first audio signal of a preset duration is obtained, where the first audio signal includes at least one audio frame signal.
Multiple wavelet decomposition signals respectively corresponding to the at least one audio frame signal are obtained by performing wavelet decomposition on each audio frame signal, where each wavelet decomposition signal contains multiple sample points and an audio intensity value of each sample point.
A wavelet signal sequence is obtained by combining the wavelet decomposition signals corresponding to the at least one audio frame signal according to a framing sequence of the at least one audio frame signal in the first audio signal. A maximum value and a minimum value among audio intensity values of all sample points in the wavelet signal sequence are obtained, and a first audio intensity threshold is determined according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence.
Sample points each having an audio intensity value greater than the first audio intensity threshold in the wavelet signal sequence are obtained, and a signal of sample points in the first audio signal corresponding to the sample points each having an audio intensity value greater than the first audio intensity threshold in the wavelet signal sequence is determined as the valid voice signal.
In some possible embodiments, the first audio intensity threshold is determined according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence as follows.
The first audio intensity threshold and a second audio intensity threshold are determined according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, where the first audio intensity threshold is less than the second audio intensity threshold.
The signal of the sample points in the first audio signal corresponding to the sample points each having the audio intensity value greater than the first audio intensity threshold in the wavelet signal sequence is determined as the valid voice signal as follows.
A first sample point in the wavelet signal sequence is obtained, where an audio intensity value of a sample point previous to the first sample point is less than the second audio intensity threshold, and an audio intensity value of the first sample point is greater than the second audio intensity threshold.
A second sample point in the wavelet signal sequence is obtained, where the second sample point is after the first sample point and is the first of sample points each having an audio intensity value less than the first audio intensity threshold in the wavelet signal sequence.
A signal of the sample points in the first audio signal corresponding to sample points from the first sample point to a sample point previous to the second sample point in the wavelet signal sequence is determined as a valid voice segment in the valid voice signal.
Optionally, at least a first preset number of consecutive sample points are included between the second sample point and the first sample point.
In some possible embodiments, the method further includes the following. An average value of first reference audio intensity values of a second preset number of consecutive sample points including a target sample point in the wavelet signal sequence is determined as an audio intensity value of the target sample point.
In some possible embodiments, prior to determining the average value of the first reference audio intensity values of the second preset number of consecutive sample points including the target sample point in the wavelet signal sequence as the audio intensity value of the target sample point, the method further includes the following.
A second reference audio intensity value of the target sample point is obtained by multiplying an audio intensity value of a sample point previous to the target sample point in the wavelet signal sequence by a smoothing coefficient.
A third reference audio intensity value of the target sample point is obtained by multiplying an average value of audio intensity values of sample points that include the target sample point and all consecutive sample points previous to the target sample point in the wavelet signal sequence by a remaining smoothing coefficient.
A sum of the second reference audio intensity value and the third reference audio intensity value is determined as a fourth reference audio intensity value of the target sample point. A minimum value among fourth reference audio intensity values of sample points including the target sample point and all sample points previous to the target sample point in the wavelet signal sequence is determined as the first reference audio intensity value of the target sample point.
In some possible implementations, the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence are obtained as follows.
A value obtained by processing a reference maximum value of each of all the wavelet decomposition signals in the wavelet signal sequence is determined as the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, and a value obtained by processing a reference minimum value of each of all the wavelet decomposition signals in the wavelet signal sequence is determined as the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence.
For each wavelet decomposition signal, the reference maximum value of the wavelet decomposition signal is obtained according to a maximum value among audio intensity values of all sample points of the wavelet decomposition signal, and the reference minimum value of the wavelet decomposition signal is obtained according to a minimum value among audio intensity values of all sample points of the wavelet decomposition signal.
Optionally, prior to obtaining the first audio signal of the preset duration, the method further includes the following. The first audio signal is obtained by compensating for a high-frequency component in an original audio signal of the preset duration.
In some possible implementations, the wavelet decomposition is performed on each audio frame signal as follows. Wavelet packet decomposition is performed on each audio frame signal, and each signal obtained after the wavelet packet decomposition is determined as the wavelet decomposition signal.
In some possible embodiments, a first audio intensity threshold is determined according to TL=min(λ1·(Scmax−Scmin)+Scmin, λ2·Scmin) and the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence. Scmax represents the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, Scmin represents the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, λ1 represents a second preset threshold, and λ2 represents a third preset threshold.
In some possible implementations, the first audio intensity threshold and the second audio intensity threshold are determined according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence as follows.
A first audio intensity threshold is determined according to TL=min(λ1·(Scmax−Scmin)+Scmin, λ2·Scmin) and the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence. Scmax represents the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, Scmin represents the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, λ1 represents a second preset threshold, and λ2 represents a third preset threshold.
The second audio intensity threshold is determined according to TU=αTL, where α represents a fourth preset threshold and is greater than 1.
According to a second aspect, a device for detecting a voice signal is provided in the disclosure, which includes an obtaining module, a decomposition module, a combining module, and a determining module.
The obtaining module is configured to obtain a first audio signal of a preset duration, where the first audio signal includes at least one audio frame signal.
The decomposition module is configured to obtain multiple wavelet decomposition signals respectively corresponding to the at least one audio frame signal by performing wavelet decomposition on each audio frame signal, where each wavelet decomposition signal contains multiple sample points and an audio intensity value of each sample point.
The combining module is configured to obtain a wavelet signal sequence by combining the wavelet decomposition signals corresponding to the at least one audio frame signal according to a framing sequence of the at least one audio frame signal in the first audio signal.
The determining module is configured to obtain a maximum value and a minimum value among audio intensity values of all sample points in the wavelet signal sequence, and determine a first audio intensity threshold according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence.
The determining module is further configured to obtain sample points each having an audio intensity value greater than the first audio intensity threshold in the wavelet signal sequence, and determine a signal of sample points in the first audio signal corresponding to the sample points each having an audio intensity value greater than the first audio intensity threshold in the wavelet signal sequence as the valid voice signal.
In some possible embodiments, the determining module is further configured to determine the first audio intensity threshold and a second audio intensity threshold according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, where the first audio intensity threshold is less than the second audio intensity threshold.
The obtaining module is further configured to obtain a first sample point in the wavelet signal sequence, where an audio intensity value of a sample point previous to the first sample point is less than the second audio intensity threshold, and an audio intensity value of the first sample point is greater than the second audio intensity threshold.
The obtaining module is further configured to obtain a second sample point in the wavelet signal sequence, where the second sample point is after the first sample point and is the first of sample points each having an audio intensity value less than the first audio intensity threshold in the wavelet signal sequence.
The determining module is further configured to determine a signal of the sample points in the first audio signal corresponding to sample points from the first sample point to a sample point previous to the second sample point in the wavelet signal sequence as a valid voice segment in the valid voice signal.
Optionally, at least a first preset number of consecutive sample points are included between the second sample point and the first sample point.
In some possible embodiments, the determining module is further configured to determine an average value of first reference audio intensity values of a second preset number of consecutive sample points including a target sample point in the wavelet signal sequence as an audio intensity value of the target sample point.
In some possible implementations, the device for detecting a voice signal further includes a calculating module. Before the determining module determines the average value of the first reference audio intensity values of the second preset number of consecutive sample points including the target sample point in the wavelet signal sequence as the audio intensity value of the target sample point, the calculating module is configured to obtain a second reference audio intensity value of the target sample point by multiplying an audio intensity value of a sample point previous to the target sample point in the wavelet signal sequence by a smoothing coefficient.
The calculating module is further configured to obtain a third reference audio intensity value of the target sample point by multiplying an average value of audio intensity values of sample points that include the target sample point and all consecutive sample points previous to the target sample point in the wavelet signal sequence by a remaining smoothing coefficient.
The calculating module is further configured to determine a sum of the second reference audio intensity value and the third reference audio intensity value as a fourth reference audio intensity value of the target sample point.
The determining module is further configured to determine a minimum value among fourth reference audio intensity values of sample points including the target sample point and all sample points previous to the target sample point in the wavelet signal sequence as the first reference audio intensity value of the target sample point.
Optionally, the determining module is further configured to determine a value obtained by processing a reference maximum value of each of all the wavelet decomposition signals in the wavelet signal sequence as the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, and determine a value obtained by processing a reference minimum value of each of all the wavelet decomposition signals in the wavelet signal sequence as the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence.
For each wavelet decomposition signal, the reference maximum value of the wavelet decomposition signal is obtained according to a maximum value among audio intensity values of all sample points of the wavelet decomposition signal, and the reference minimum value of the wavelet decomposition signal is obtained according to a minimum value among audio intensity values of all sample points of the wavelet decomposition signal.
In some possible embodiments, the device 14 for detecting a voice signal further includes a compensating module. Before the obtaining modules obtains the first audio signal of the preset duration, the compensating module is configured to obtain the first audio signal by compensating for a high-frequency component in an original audio signal of the preset duration.
In some possible implementations, the decomposition module is further configured to perform wavelet packet decomposition on each audio frame signal, and determine each signal obtained after the wavelet packet decomposition as the wavelet decomposition signal.
In some possible implementations, the determining module is further configured to determine a first audio intensity threshold according to TL=min(λ1·(Scmax−Scmin)+Scmin, λ2·Scmin) and the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence. Scmax represents the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, Scmin represents the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, λ1 represents a second preset threshold, and λ2 represents a third preset threshold.
In some possible implementations, the determining module is further configured to determine a first audio intensity threshold according to TL=min(λ1·(Scmax−Scmin)+Scmin, λ2. Scmin) and the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, where Scmax represents the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, Scmin represents the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, λ1 represents a second preset threshold, and λ2 represents a third preset threshold; and determine the second audio intensity threshold according to TU=αTL, where α represents a fourth preset threshold and is greater than 1.
According to a third aspect, an apparatus for detecting a valid voice signal is provided in the disclosure. The apparatus includes a receiver, a processor, and a memory.
The transceiver is coupled with the processor and the memory, and the processor is further coupled with the memory.
The transceiver is configured to obtain a first audio signal of a preset duration, where the first audio signal includes at least one audio frame signal.
The processor is configured to obtain multiple wavelet decomposition signals respectively corresponding to the at least one audio frame signal by performing wavelet decomposition on each audio frame signal, where each wavelet decomposition signal contains multiple sample points and an audio intensity value of each sample point.
The processor is further configured to obtain a wavelet signal sequence by combining the wavelet decomposition signals corresponding to the at least one audio frame signal according to a framing sequence of the at least one audio frame signal in the first audio signal.
The processor is further configured to obtain a maximum value and a minimum value among audio intensity values of all sample points in the wavelet signal sequence, and determine a first audio intensity threshold according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence.
The processor is further configured to obtain sample points each having an audio intensity value greater than the first audio intensity threshold in the wavelet signal sequence, and determine a signal of sample points in the first audio signal corresponding to the sample points each having an audio intensity value greater than the first audio intensity threshold in the wavelet signal sequence as the valid voice signal.
The memory is configured to store computer programs, and the computer programs are invoked by the processor.
In some possible embodiments, the processor is further configured to: determine the first audio intensity threshold and a second audio intensity threshold according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, where the first audio intensity threshold is less than the second audio intensity threshold; obtain a first sample point in the wavelet signal sequence, where an audio intensity value of a sample point previous to the first sample point is less than the second audio intensity threshold, and an audio intensity value of the first sample point is greater than the second audio intensity threshold; obtain a second sample point in the wavelet signal sequence, where the second sample point is after the first sample point and is the first of sample points each having an audio intensity value less than the first audio intensity threshold in the wavelet signal sequence; and determine a signal of the sample points in the first audio signal corresponding to sample points from the first sample point to a sample point previous to the second sample point in the wavelet signal sequence as a valid voice segment in the valid voice signal.
Optionally, at least a first preset number of consecutive sample points are included between the second sample point and the first sample point.
In some possible embodiments, the processor is further configured to determine an average value of first reference audio intensity values of a second preset number of consecutive sample points including a target sample point in the wavelet signal sequence as an audio intensity value of the target sample point.
In some possible embodiments, the processor is further configured to obtain a second reference audio intensity value of the target sample point by multiplying an audio intensity value of a sample point previous to the target sample point in the wavelet signal sequence by a smoothing coefficient; obtain a third reference audio intensity value of the target sample point by multiplying an average value of audio intensity values of sample points that include the target sample point and all consecutive sample points previous to the target sample point in the wavelet signal sequence by a remaining smoothing coefficient; determine a sum of the second reference audio intensity value and the third reference audio intensity value as a fourth reference audio intensity value of the target sample point; and determine a minimum value among fourth reference audio intensity values of sample points including the target sample point and all sample points previous to the target sample point in the wavelet signal sequence as the first reference audio intensity value of the target sample point.
In some possible implementations, the processor is further configured to determine a value obtained by processing a reference maximum value of each of all the wavelet decomposition signals in the wavelet signal sequence as the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, and determine a value obtained by processing a reference minimum value of each of all the wavelet decomposition signals in the wavelet signal sequence as the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, where for each wavelet decomposition signal, the reference maximum value of the wavelet decomposition signal is obtained according to a maximum value among audio intensity values of all sample points of the wavelet decomposition signal, and the reference minimum value of the wavelet decomposition signal is obtained according to a minimum value among audio intensity values of all sample points of the wavelet decomposition signal.
Optionally, the processor is further configured to obtain the first audio signal by compensating for a high-frequency component in an original audio signal of the preset duration.
In some possible embodiments, the processor is further configured to perform wavelet packet decomposition on each audio frame signal, and determine each signal obtained after the wavelet packet decomposition as the wavelet decomposition signal.
In some possible implementations, the processor is configured to determine a first audio intensity threshold according to TL=min(λ1·(Scmax−Scmin)+Scmin, λ2·Scmin) and the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence. Scmax represents the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, Scmin represents the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, λ1 represents a second preset threshold, and λ2 represents a third preset threshold.
In some possible implementations, the processor is configured to determine a first audio intensity threshold according to TL=min(λ1·(Scmax−Scmin)+Scmin, λ2·Scmin) and the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, where Scmax represents the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, Scmin represents the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, λ1 represents a second preset threshold, and λ2 represents a third preset threshold; and determine the second audio intensity threshold according to TU=αTL, where α represents a fourth preset threshold and is greater than 1.
According to a fourth aspect, a non-transitory computer readable storage medium is provided in the disclosure. The readable storage medium stores instructions which, when executed on a computer, can perform operations of the method described in the above aspect.
Technical solutions embodied in embodiments of the disclosure will be described in a clear and comprehensive manner in conjunction with accompanying drawings of the embodiments of the disclosure. It is evident that the embodiments described herein are some rather than all the embodiments of the disclosure. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the disclosure without creative efforts shall fall within the protection scope of the disclosure.
Implementations of the technical solutions of the disclosure will be further described in detail below in conjunction with accompanying drawings.
Referring to
Referring to
At 100, a first audio signal of a preset duration is obtained, where the first audio signal includes at least one audio frame signal. Specifically, a device for detecting a valid voice signal obtains the first audio signal of the preset duration. Since movement of oral muscles is relatively slow relative to a voice frequency, and voice signal is relatively stable in a short time range, the voice signal has short-term stability. Therefore, the voice signal can be segmented into segments for detection according to the short-term stability of the voice signal. That is, framing is performed on the first audio signal of the preset duration to obtain at least one audio frame signal. Optionally, there is no overlap between audio frame signals, and a frame shift is the same as a frame length. It can be understood that the frame shift can be regarded as an overlap between a previous frame and a next frame. When the frame length equals the frame shift, there is no overlap between audio frames. In some possible embodiments, the device for detecting a valid voice signal samples the voice signal at a frequency of 16 kHz, i.e., collects 16 k sample points in one second. Thereafter, a first audio signal with the preset duration of 5 seconds is obtained, and then framing is performed on the first audio signal with 10 ms as the frame shift and 10 ms as the frame length. Therefore, each audio frame signal includes 160 sample points, and an audio intensity value of each of the 160 sample points is obtained.
At 101, multiple wavelet decomposition signals respectively corresponding to the at least one audio frame signal are obtained by performing wavelet decomposition on each audio frame signal, where each wavelet decomposition signal contains multiple sample points and an audio intensity value of each sample point. Specifically, at 100, the first audio signal is obtained, framing is performed on the first audio signal to obtain audio frame signals, and then the wavelet decomposition is performed on each audio frame signal.
The wavelet decomposition is illustrated in detail below. The wavelet decomposition is illustrated in
Specific processing of the wavelet decomposition in embodiments is illustrated in detail below. In embodiments, take performing the wavelet decomposition on one audio frame signal as an example for illustration. Specifically, referring to
In some possible embodiments, the wavelet decomposition is performed on each audio frame signal as follows. Wavelet packet decomposition is performed on each audio frame signal, and each signal obtained after the wavelet packet decomposition is determined as the wavelet decomposition signal.
The wavelet packet decomposition is illustrated in detail below. The wavelet packet decomposition may be illustrated in
Specific processing of the wavelet packet decomposition in embodiments is illustrated in detail below. In embodiments, take performing the wavelet packet decomposition on one audio frame signal as an example for illustration. Specifically, referring to
At 102, a wavelet signal sequence is obtained by combining the wavelet decomposition signals corresponding to the at least one audio frame signal according to a framing sequence of the at least one audio frame signal in the first audio signal. Specifically, the wavelet decomposition signal of the first audio frame signal is obtained according to operations at 101, wavelet decomposition signals of all audio frame signals in the first audio signal are obtained, and then the wavelet decomposition signals of all the audio frame signals are sequentially combined according to the framing sequence of the first audio signal described at 100 to obtain the wavelet signal sequence representing information of the first audio signal.
At 103, a maximum value and a minimum value among audio intensity values of all sample points in the wavelet signal sequence are obtained, and a first audio intensity threshold is determined according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence. Specifically, for each of all the sample points in the wavelet signal sequence, a sample point value of the sample point represents a voltage amplitude value of the sample point. In some possible implementations, the audio intensity value may be the voltage amplitude value of the sample point. In other possible implementations, the audio intensity value may be an energy value of the sample point. The energy value of the sample point is obtained by squaring the voltage amplitude value of the sample point. The first audio intensity threshold which is used for determination of the valid voice signal is determined according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence. In some possible implementations, the device for detecting a valid voice signal determines the first audio intensity threshold TL according to TL=min(λ1·(Scmax−Scmin)+Scmin, λ2·Scmin) and the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence. Scmax represents the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, Scmin represents the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, λ1 represents a second preset threshold, and λ2 represents a third preset threshold. In one example, λ1 is 0.04, and λ2 is 50.
In some possible embodiments, the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence are obtained as follows. A first reference maximum value and a first reference minimum value among audio intensity values of all sample points of a first wavelet decomposition signal in the wavelet signal sequence are obtained. A value obtained by processing a reference maximum value of each of all the wavelet decomposition signals in the wavelet signal sequence is determined as the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, where for each wavelet decomposition signal, the reference maximum value of the wavelet decomposition signal is obtained according to a maximum value among audio intensity values of all sample points of the wavelet decomposition signal. A value obtained by processing a reference minimum value of each of all the wavelet decomposition signals in the wavelet signal sequence is determined as the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, where for each wavelet decomposition signal, the reference minimum value of the wavelet decomposition signal is obtained according to a minimum value among audio intensity values of all sample points of the wavelet decomposition signal. Specifically, the wavelet signal sequence includes multiple wavelet decomposition signals, and a maximum value and a minimum value of all sample points of each wavelet decomposition signal are obtained. Optionally, an average value of the maximum values in all the wavelet decomposition signals is determined as the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence. An average value of the minimum values in all the wavelet decomposition signals is determined as the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence. According to the embodiments, the maximum value and the minimum value in the wavelet signal sequence are optimized, such that the sample points in the wavelet signal sequence can be further analyzed to optimize detection effect of the valid voice signal.
At 104, sample points each having an audio intensity value greater than the first audio intensity threshold in the wavelet signal sequence are obtained, and a signal of sample points in the first audio signal corresponding to the sample points each having an audio intensity value greater than the first audio intensity threshold in the wavelet signal sequence is determined as the valid voice signal.
In some possible embodiments, prior to obtaining the first audio signal of the preset duration, the method further includes the following. The first audio signal is obtained by compensating for a high-frequency component in an original audio signal of the preset duration. Specifically, due to loss of high-frequency components in voice signal during lip pronunciation or microphone recording and great loss of the signal during transmission with the increase of a signal rate, to obtain a relatively good signal waveform at a receiving terminal, the loss signal needs to be compensated. In some possible implementations, the original audio signal of the preset duration is pre-emphasized. That is, the first audio signal is processed according to y(n)=x(n)−ax(n−1), where x(n) represents an audio intensity value of a sample point of the original audio signal at time n, x(n−1) represents an audio intensity value of a sample point of the original audio signal at time n−1, a represents a pre-emphasis coefficient (for example, a is greater than 0.9 and less than 1), which can be deemed as the first preset threshold, and y(n) represents a signal subjected to pre-emphasizing. It can be understood that, it can be considered that the pre-emphasizing is to compensate for high-frequency components by passing the first audio signal through a high-pass filter, such that the loss of the high-frequency components caused by lip articulation or microphone recording can be reduced.
According to embodiments, by collecting the energy information of all the sample points in the wavelet signal sequence, the audio intensity threshold is determined according to the energy distribution of the wavelet signal sequence, and determination and detection of the valid voice signal may be realized according to the audio intensity threshold, thereby improving the accuracy of the detection of the valid voice signal.
The following may describe another method for detecting a valid voice signal provided in the disclosure with reference to
Referring to
At 700, a first audio signal of a preset duration is obtained, where the first audio signal includes at least one audio frame signal.
At 701, multiple wavelet decomposition signals respectively corresponding to the at least one audio frame signal are obtained by performing wavelet decomposition on each audio frame signal, where each wavelet decomposition signal contains multiple sample points and an audio intensity value of each sample point.
At 702, a wavelet signal sequence is obtained by combining the wavelet decomposition signals corresponding to the at least one audio frame signal according to a framing sequence of the at least one audio frame signal in the first audio signal.
It can be understood that operations at 700, 701, and 702 correspond to performing framing on the first audio signal and obtaining the wavelet signal sequence by combining signals obtained after wavelet decomposition. For specific implementations, reference may be made to embodiments described above in conjunction with
At 703, a maximum value and a minimum value among audio intensity values of all sample points in the wavelet signal sequence are obtained, and a first audio intensity threshold and a second audio intensity threshold are determined according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, where the first audio intensity threshold is less than the second audio intensity threshold. Specifically, the first audio intensity threshold and the second audio intensity threshold are determined according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence. Optionally, the first audio intensity threshold TL is determined according to TL=min(λ1·(Scmax−Scmin)+Scmin, λ2·Scmin) and the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence. Scmax represents the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, Scmin represents the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, λ1 represents a second preset threshold, and λ2 represents a third preset threshold. The second audio intensity threshold TU is determined according to TU=αTL, where α represents a fourth preset threshold and is greater than 1.
In some possible embodiments, the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence are obtained as follows. A value obtained by processing a reference maximum value of each of all the wavelet decomposition signals in the wavelet signal sequence is determined as the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, and a value obtained by processing a reference minimum value of each of all the wavelet decomposition signals in the wavelet signal sequence is determined as the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence. For each wavelet decomposition signal, the reference maximum value of the wavelet decomposition signal is obtained according to a maximum value among audio intensity values of all sample points of the wavelet decomposition signal, and the reference minimum value of the wavelet decomposition signal is obtained according to a minimum value among audio intensity values of all sample points of the wavelet decomposition signal. Specifically, the wavelet signal sequence includes multiple wavelet decomposition signals, and a maximum value and a minimum value of all sample points of each wavelet decomposition signal are obtained. Optionally, an average value of the maximum values in all the wavelet decomposition signals is determined as the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence. An average value of the minimum values in all the wavelet decomposition signals is determined as the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence. According to the embodiments, the maximum value and the minimum value in the wavelet signal sequence are optimized, such that the sample points in the wavelet signal sequence can be further analyzed to optimize detection effect of the valid voice signal.
At 704, a first sample point in the wavelet signal sequence is obtained, where an audio intensity value of a sample point previous to the first sample point is less than the second audio intensity threshold, and an audio intensity value of the first sample point is greater than the second audio intensity threshold. Specifically, when the audio intensity value of the first sample point in the wavelet signal sequence is greater than the second audio intensity threshold, and the audio intensity value of the sample point previous to the first sample point is less than the second audio intensity threshold, the first sample point may be deemed as a starting point of the valid voice signal, that is, it is predefined to enter a valid voice segment from the first sample point.
At 705, a second sample point in the wavelet signal sequence is obtained, where the second sample point is after the first sample point and is the first of sample points each having an audio intensity value less than the first audio intensity threshold in the wavelet signal sequence. Specifically, the first sample point is predefined as the starting point of the valid voice segment at 704, i.e., entering the valid voice segment from the first sample point. After the first sampling point, when the second sample point is the first of sample points each having the audio intensity value less than the first audio intensity threshold, it can be considered that the second sample point has exited the valid voice segment in which the first sample point is located.
At 706, a signal of the sample points in the first audio signal corresponding to sample points from the first sample point to a sample point previous to the second sample point in the wavelet signal sequence is determined as a valid voice segment in the valid voice signal. Specifically, the second sample point has exited the valid voice segment in which the first sample point is located at 705, and thus it can be determined that the signal of the sample points in the first audio signal corresponding to the sample points from the first sample point to the sample point previous to the second sample point can be determined as the valid voice segment. In addition, at least a first preset number of consecutive sample points are included between the second sample point and the first sample point. If the first sample point and the second sample point are too close, for example, the first preset number is 20 and the number of consecutive sample points between the first sample point and the second sample point is less than the first preset number, it can be considered that the audio intensity value of the first sample point being greater than the second audio intensity threshold is caused by jitter of transient noise rather than by valid voice.
In some possible embodiments, prior to obtaining the first audio signal of the preset duration, the method further includes the following. The first audio signal is obtained by compensating for a high-frequency component in an original audio signal of the preset duration. Specifically, due to loss of high-frequency components in voice signal during lip pronunciation or microphone recording and great loss of the signal during transmission with the increase of a signal rate, to obtain a relatively good signal waveform at a receiving terminal, the loss signal needs to be compensated. In some possible implementations, the original audio signal of the preset duration is pre-emphasized. That is, the first audio signal is processed according to y(n)=x(n)−ax(n−1), where x(n) represents an audio intensity value of a sample point of the original audio signal at time n, x(n−1) represents an audio intensity value of a sample point of the original audio signal at time n−1, a represents a pre-emphasis coefficient (for example, a is greater than 0.9 and less than 1), which can be deemed as the first preset threshold, and y(n) represents a signal subjected to pre-emphasizing. It can be understood that, it can be considered that the pre-emphasizing is to compensate for high-frequency components by passing the first audio signal through a high-pass filter, such that the loss of the high-frequency components caused by lip articulation or microphone recording can be reduced.
The effect of implementing the embodiments is illustrated in
Specific implementations of embodiments are described in detail with reference to the accompanying drawings. Referring to
At 900, a device for detecting a valid signal initially defines a sample point index i=0, a valid voice signal starting point index is=0, and a valid voice signal time period index idx=0. Specifically, the sample point index i is an independent variable, which represents an ith sample point. The starting point index (is) is a recording variable, which records a starting sample point of the valid signal segment. To traverse all the sample points in the wavelet signal sequence, the independent variable i may change, and so it is necessary to define the variable is to record a first same point. Optionally, the valid voice signal time period index idx is also a recording variable, which records a (idx)th valid voice segment. idx may be defined to record the number of valid voice segments included in the first audio signal.
At 901, whether an audio intensity value Sc(i) of the ith sample point is greater than the second audio intensity threshold and the starting point index (is) is equal to 0 are determined. Specifically, the second audio intensity threshold can be considered as an upper limit threshold of the valid voice signal, and the audio intensity value of the sample point is compared with the second audio intensity threshold.
At 902, the sample point i which represents entering a valid voice segment is recorded (i.e., is=i). Specifically, when the audio intensity value Sc(i) of the ith sample point is greater than the second audio intensity threshold and the starting point index (is) is equal to 0 as initially defined, a sample point position of the ith sample point determined at 901 is recorded by is, and it is predefined to enter the valid voice segment from the ith sample point. Thereafter, an audio intensity value of a next sample point is compared. That is, proceed to operations at 907 (i=i+1). In other words, the next sample point is taken as a present sample point, to continue detection and determination. It can be understood that the audio intensity value of the sample point previous to the ith sample point is less than the second audio intensity threshold and the audio intensity value of the ith sample point is greater than the second audio intensity threshold. A first sample point in the wavelet signal sequence is obtained according to operations at 704 of embodiments described above in conjunction with
At 903, whether the audio intensity value Sc(i) of the ith sample point is less than the second audio intensity threshold and the starting point index (is) is not 0 are determined. Specifically, if the audio intensity value Sc(i) of the ith sample point is less than or equal to the second audio intensity threshold or the starting point index (is) is not 0, the audio intensity value Sc(i) of the ith sample point is compared with the first audio intensity threshold, to obtain a second sample point in the wavelet signal sequence according to operations at 705 of embodiments described above in conjunction with
Furthermore, a time interval between the starting sample point entering the valid voice signal segment and the end sample point of the valid voice signal may be compared to determine whether at least a first preset number of consecutive sample points are included between the first sample point and the second sample point, which are described as follows.
At 904, whether a time interval between i and is is greater than Tmin, i.e., i>is +Tmin is determined. Specifically, since a sampling interval can be determined according to a sampling frequency, at least the first preset number of consecutive sample points is included between the first sample point and the second sample point, and the first preset number of consecutive sample points can be represented by a time period Tmin. For example, 16 kHz is taken as an example of the sampling frequency of the first audio frame signal, a frame length of the first audio frame signal is 10 ms, and the first audio frame signal includes 160 sample points. After down-sampling of three-stage wavelet decomposition or three-stage wavelet packet decomposition is performed, an interval between sampling points in the wavelet signal sequence is 0.5 ms. If the first preset number is 20, Tmin equals 20 multiplied by 0.5 ms, that is, Tmin equals 10 ms. If at least the first preset number of consecutive sample points are included between the first sample point and the second sample point, i.e., i>is+Tmin, proceed to operations at 905. If the number of sample points between the first sample point and the second sample point is less than the first preset number of consecutive sample points, i is the second sample point, and is is the first sample point determined at 901, i>is+Tmin is not true, and it can be considered that a (i−1)th sample point previous to the ith sample point is not the end sample point of the valid voice segment. The starting sample point of the valid voice segment recorded by is=i at 902 may be caused by noise jitter. Since energy of transient noise rapidly rises and then rapidly falls, the audio intensity value of the sample point may be greater than the second audio intensity threshold in a short time period, and then may drop below the first audio intensity threshold in a time period less than Tmin, which is inconsistent with the short-term stability of the voice signal. Therefore, the signal segment is discarded, and proceed to operations at 906.
At 905, idx=idx+1, and the valid voice segment is [is, i−1]. Specifically, when at least the first preset number of consecutive sample points are included between the first sample point and the second sample point, i.e., i>is+Tmin, a signal of the sample points in the first audio signal corresponding to sample points from the first sample point to a sample point previous to the second sample point in the wavelet signal sequence is determined as a valid voice segment in the valid voice signal according to operations at 706 in the implementations described above in combination with
At 906, reset is=0. Specifically, since the first sample point recorded by is has been recorded in the interval, a value of is can be released, i.e., reset is=0, and proceed to operations at 907 (i=i+1). In other words, a next sample point may be taken as the present sample point to perform detection of another valid voice segment.
At 907, i=i+1. Specifically, continue to traverse sample points in the wavelet signal sequence, i.e., sequentially traverse the sample points by increasing i by one.
At 908, whether i is greater than or equal to the total number of sample points is determined. Specifically, after operations at 907 (i=i+1) is performed, before another valid voice segment is detected, it is necessary to determine a position of the sample point, i.e., determine whether i of the ith sample point is greater than or equal to the total number of sample points in the wavelet signal sequence, because i is kept increasing by one to sequentially traverse subsequent sample points. If i is less than the total number of sample points in the wavelet signal sequence, proceed to comparing the audio intensity value with the second audio intensity threshold or the first audio intensity threshold. If the ith sample point has been traversed to the last of all the sample points, e.g., i is equal to or greater than the total number of sample points, proceed to operations at 909.
At 909, determine that the valid voice segment is [is, i−1], to determine, according to operations at 706 in embodiments described above in conjunction with
In foregoing embodiments described in conjunction with
Referring to
At 1000, a second reference audio intensity value of a target sample point is obtained by multiplying an audio intensity value of a sample point previous to the target sample point in the wavelet signal sequence by a smoothing coefficient. Specifically, time-domain amplitude smoothing is performed on the sample points in the wavelet signal sequence, to enable a smooth transition between adjacent sample points in the voice signal, thereby reducing the influence of the burr on the voice signal. In one example, if S(i) represents the audio intensity value of the target sample point, S(i−1) represents the audio intensity value of the sample point previous to the target sample point, and αs represents the smoothing coefficient, the audio intensity value S(i−1) of the sample point previous to the target sample point in the wavelet signal sequence is multiplied by the smoothing coefficient αs to obtain the second reference audio intensity value of the target sample point. The second reference audio intensity value of the target sample point may be expressed by αs×S(i−1).
At 1001, a third reference audio intensity value of the target sample point is obtained by multiplying an average value of audio intensity values of sample points that include the target sample point and all consecutive sample points previous to the target sample point in the wavelet signal sequence by a remaining smoothing coefficient. Specifically, the second reference audio intensity value is determined as a part of a time-domain smoothing result. A value obtained by multiplying the average value of the audio intensity values of the sample points that include the target sample point and all the consecutive sample points previous to the target sample point in the wavelet signal sequence by the remaining smoothing coefficient is determined as the other part of the time-domain smoothing result. In one example, take performing three-stage wavelet packet decomposition on the first audio signal as an example for illustration. The wavelet signal sequence includes eight wavelet packet decomposition signals. The average value M(i) of the audio intensity values of all the consecutive sample points previous to the target sample point can be expressed as:
In formula 1, i represents the ith sample point in the wavelet signal sequence, and l represents a lth wavelet decomposition signal. It can be understood that i is less than the total number of all sample points in the wavelet signal sequence. The third reference audio intensity value of the target sample point is obtained by multiplying the average value M(i) of the audio intensity values of the sample points that include the target sample point and all the consecutive sample points previous to the target sample point in the wavelet signal sequence by the remaining smoothing coefficient 1−αs. The third reference audio intensity value can be expressed by M(i)×(1−αs).
At 1002, a sum of the second reference audio intensity value and the third reference audio intensity value is determined as a fourth reference audio intensity value of the target sample point. Specifically, according to operations at 1000, it can be obtained that the second reference audio intensity value is αs×S(i−1), and according to operations at 1001, it can be obtained that the third reference audio intensity value is M(i)×(1−αs). Therefore, the fourth reference audio intensity value αs×S(i−1)+M(i)×(1−αs) is obtained by adding the second reference audio intensity value and the third reference audio intensity value. In some possible implementations, the fourth reference audio intensity value can represent the audio intensity value of the target sample point after smoothing, and then the fourth reference audio intensity value can be determined as the audio intensity value of the target sample point, which may be expressed by S(i)=αs×S(i−1)+M(i)×(1−αs).
At 1003, a minimum value among fourth reference audio intensity values of sample points including the target sample point and all sample points previous to the target sample point in the wavelet signal sequence is determined as the first reference audio intensity value of the target sample point. Specifically, a duration of a signal to be tracked may be preset, and the signal of the preset duration is then segmented into tracking signals each having a first preset duration. According to the fourth reference audio intensity values of all the sample points previous to the target sample point in the wavelet signal sequence, a minimum value among fourth reference audio intensity values of all sample points in a first duration is recorded, and is passed to a tracking signal of a next preset duration. That is, the minimum value of all the sample points in the previous preset duration is compared with an audio intensity value of a first sample point in a present preset duration and then the smaller of the two values is recorded. Thereafter, the smaller of the two values is compared with an audio intensity value of a subsequent sample point in the present preset duration. In above manners, the smaller of the two values is recorded each time and is then compared with an audio intensity value of a subsequent sample point. Therefore a minimum value among fourth reference audio intensity values of all the sample points in the preset duration is obtained, such that a first reference audio intensity value of the target sample point can be determined.
By implementing embodiments, all the sample points in the wavelet signal sequence are segmented according to a preset duration, and the distribution of audio intensity of all the sample points in the preset duration is tracked, so that the energy of transient noise can be weakened. The effect of implementing implementations can be illustrated in
The following may describe in detail how to track the voice signal and the effect achieved by tracking the voice signal with reference to the accompanying drawings.
In some possible embodiments, to further reduce the influence of blur that may appear in the wavelet signal sequence, after the first reference audio intensity value of the target sample point is determined, the following can be further conducted.
At 1004, an average value of first reference audio intensity values of a second preset number of consecutive sample points including the target sample point in the wavelet signal sequence is determined as an audio intensity value of the target sample point. Specifically, in the wavelet signal sequence, short-term mean smoothing is performed on the target sample point, and a value obtained after the short-term mean smoothing is determined as the audio intensity value of the target sample point. In some possible implementations, an audio intensity value SC(i) of a ith sample point is:
In formula 2, 2M represents the second preset number of consecutive sample points, Sm(i) represents the first reference audio intensity value of the target sample point, and Sm(i−m) represents m sample points before or after the ith sample point. In some examples, M=80, i.e., the second preset number of consecutive sample points is 160, and therefore Σm=−MM Sm(i−m) may represent that a sum operation is performed on first reference audio intensity values of 80 sample point before the ith sample point and first reference audio intensity values of 80 sample points after the ith sample point, to obtain a sum of the audio intensity value of each of sample points that include target sample point i, M sample points before target sample point i, and M sample points after target sample point i. Thereafter, a result obtained after the sum operation is averaged. That is, the sum of the audio intensity values is divided by the number of all sample points, and a result obtained after dividing is then determined as the audio intensity value SC(i) of the ith sample point after amplitude short-term mean smoothing. In the formula 2, m is an independent variable. To avoid negative sample points, i is greater than M. M being equal to 80 is taken as an example, i.e., mean smoothing is performed on sample points starting from a 81st sample point.
The device for detecting a valid voice signal tracks the voice signal and uses the tracking result to affect the audio intensity value of the signal, which can be combined with any one of the implementations described above in conjunction with
In some possible embodiments, the device for detecting a valid voice signal obtains a first audio signal of a preset duration, and obtains multiple sample points of each audio frame signal and an audio intensity value of each sample point, where the first audio signal includes at least one audio frame signal.
Multiple wavelet decomposition signals respectively corresponding to the at least one audio frame signal are obtained by performing wavelet decomposition on each audio frame signal, where each wavelet decomposition signal contains multiple sample points and an audio intensity value of each sample point.
A wavelet signal sequence is obtained by combining the wavelet decomposition signals corresponding to the at least one audio frame signal according to a framing sequence of the at least one audio frame signal in the first audio signal.
A second reference audio intensity value of the target sample point is obtained by multiplying an audio intensity value of a sample point previous to the target sample point in the wavelet signal sequence by a smoothing coefficient.
A third reference audio intensity value of the target sample point is obtained by multiplying an average value of audio intensity values of sample points that include the target sample point and all consecutive sample points previous to the target sample point in the wavelet signal sequence by a remaining smoothing coefficient.
A sum of the second reference audio intensity value and the third reference audio intensity value is determined as a fourth reference audio intensity value of the target sample point.
A minimum value among fourth reference audio intensity values of sample points including the target sample point and all sample points previous to the target sample point in the wavelet signal sequence is determined as the first reference audio intensity value of the target sample point.
An average value of first reference audio intensity values of a second preset number of consecutive sample points including a target sample point in the wavelet signal sequence is determined as an audio intensity value of the target sample point.
A maximum value and a minimum value among audio intensity values of all sample points in the wavelet signal sequence are obtained, and a first audio intensity threshold is determined according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence.
Sample points each having an audio intensity value greater than the first audio intensity threshold in the wavelet signal sequence are obtained, and a signal of sample points in the first audio signal corresponding to the sample points each having an audio intensity value greater than the first audio intensity threshold in the wavelet signal sequence is determined as the valid voice signal.
For specific implementations of embodiments, reference may be made to the embodiments described above with reference to
In other possible embodiments, a device for detecting a valid voice signal obtains a first audio signal of a preset duration, where the first audio signal includes at least one audio frame signal.
Multiple wavelet decomposition signals respectively corresponding to the at least one audio frame signal are obtained by performing wavelet decomposition on each audio frame signal, where each wavelet decomposition signal contains multiple sample points and an audio intensity value of each sample point.
A wavelet signal sequence is obtained by combining the wavelet decomposition signals corresponding to the at least one audio frame signal according to a framing sequence of the at least one audio frame signal in the first audio signal.
A second reference audio intensity value of the target sample point is obtained by multiplying an audio intensity value of a sample point previous to the target sample point in the wavelet signal sequence by a smoothing coefficient.
A third reference audio intensity value of the target sample point is obtained by multiplying an average value of audio intensity values of sample points that include the target sample point and all consecutive sample points previous to the target sample point in the wavelet signal sequence by a remaining smoothing coefficient.
A sum of the second reference audio intensity value and the third reference audio intensity value is determined as a fourth reference audio intensity value of the target sample point.
A minimum value among fourth reference audio intensity values of sample points including the target sample point and all sample points previous to the target sample point in the wavelet signal sequence is determined as the first reference audio intensity value of the target sample point.
An average value of first reference audio intensity values of a second preset number of consecutive sample points including a target sample point in the wavelet signal sequence is determined as an audio intensity value of the target sample point.
A maximum value and a minimum value among audio intensity values of all sample points in the wavelet signal sequence are obtained, and the first audio intensity threshold and a second audio intensity threshold are determined according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, where the first audio intensity threshold is less than the second audio intensity threshold.
A first sample point in the wavelet signal sequence is obtained, where an audio intensity value of a sample point previous to the first sample point is less than the second audio intensity threshold, and an audio intensity value of the first sample point is greater than the second audio intensity threshold.
A second sample point in the wavelet signal sequence is obtained, where the second sample point is after the first sample point and is the first of sample points each having an audio intensity value less than the first audio intensity threshold in the wavelet signal sequence.
A signal of the sample points in the first audio signal corresponding to sample points from the first sample point to a sample point previous to the second sample point in the wavelet signal sequence is determined as a valid voice segment in the valid voice signal. Optionally, at least a first preset number of consecutive sample points are included between the second sample point and the first sample point.
For specific implementations of embodiments, reference may be made to the embodiments described above with reference to
In some possible implementations, a value obtained by processing a reference maximum value of each of all the wavelet decomposition signals in the wavelet signal sequence is determined as the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, and a value obtained by processing a reference minimum value of each of all the wavelet decomposition signals in the wavelet signal sequence is determined as the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence. For each wavelet decomposition signal, the reference maximum value of the wavelet decomposition signal is obtained according to a maximum value among audio intensity values of all sample points of the wavelet decomposition signal, and the reference minimum value of the wavelet decomposition signal is obtained according to a minimum value among audio intensity values of all sample points of the wavelet decomposition signal.
In some possible implementations, the first audio signal is obtained by compensating for a high-frequency component in an original audio signal of the preset duration.
It can be understood that the wavelet decomposition is performed on each audio frame signal as follows. Wavelet packet decomposition is performed on each audio frame signal, and each signal obtained after the wavelet packet decomposition is determined as the wavelet decomposition signal.
An exemplary illustration of how to track the voice signal will be given below with reference to the accompanying drawings. Referring to
At 1201, a device for detecting a valid voice signal initially defines a sample point index i=0 of the wavelet signal sequence, initiates an audio intensity value S(0)=M(0), and defines a sample point accumulation index imod=0. Specifically, i=0, S(0)=M(0), and imod=0 can be deemed as an initiation state of the device for detecting a valid voice signal. The number of sample points to be traversed and an audio intensity value of each sample point are initially defined, and the sample point accumulation index is used for controlling the preset duration. When a value of the sample point accumulation index imod reaches a fixed value (i.e., Vwin), data updating is conducted to complete tracking of a signal of the preset duration.
At 1202, i=i+1 and an audio intensity value of a ith sample point is S(i)=αs×S(i−1)+M(i)×(1−αs). Specifically, start to perform tracking of the audio intensity value of the sample point (which can also be understood as tracking of the energy distribution). i=i+1 represents performing amplitude smoothing on each traversed sample point, and the audio intensity value of the ith sample point after smoothing is S(i)=αs×S(i−1)+M(i)×(1−αs). Therefore, operations at 1000, 1001 and 1002 in embodiments described above in conjunction with
At 1203, whether i is less than an accumulation sample point number Vwin is determined. Specifically, in embodiments, tracking is performed on the voice signal of a time period, so sample points need to be accumulated. The accumulation sample point number Vwin is pre-defined, and optionally Vwin=10. When sample points from 0 to a ninth sample point are traversed, operations at 1204 are performed, and when a tenth sample point is traversed, operations at 1205 are performed.
At 1204, if i is less than the accumulation sample point number Vwin, define Smin=S(i) and Smact=S(i). Specifically, traversing is conducted from a first sample point in the wavelet signal sequence to perform smoothing on the audio intensity of the sample point. When i is less than accumulation sample point number Vwin, a value of S(i) is assigned to Smin and Smact, i.e., Smin=S(i), and Smact=S(i). Proceed to operations at 1206 to record data of Smin and proceed to operations at 1207 to perform sample point accumulation. In one example, i=i+1 can be understood that the device for detecting a valid voice signal keeps tracking audio intensity values of the sample points. i being less than the accumulation number Vwin includes a case where first Vwin sample points of the first audio signal are traversed. For example, Vwin=10, when a ninth sample point is traversed, Smin=S(9), and Smact=S(9), that is, Smin and Smact record the audio intensity value of the ninth sample point.
At 1205, if i is greater than or equal to the accumulation sample point number Vwin, obtain a minimum value among audio intensity values of sample points from a (Vwin)th sample point to the ith sample point, i.e., Smin=min (Smin, S(i)) and Smact=min (Smact, S(i)). Specifically, if i is greater than or equal to the accumulation sample point number Vwin, when the (Vwin)th sample point is traversed, Vwin=10 is taken as an example for illustration, i.e., when the tenth sample point is traversed at 1203, a smaller value between an audio intensity value of the ninth sample point and an audio intensity value of the tenth sample point is assigned to Smin, i.e., Smin=min (Smin, S(10)). In other words, Smin records a value of S(9) in a step before traversing to the tenth sample point.
At 1206, define Sm(i)=Smin. Specifically, operations at 1203 in embodiments described above in conjunction with
At 1207, imod=imod+1. Specifically, during traversing of sample point i, sample point accumulation index imod is also continuously accumulated, i.e., imod=imod+1, where imod is used for controlling whether to perform data updating on matrix SW. The wavelet signal sequence is segmented into voice signals each having a preset duration for tracking. It can be understood that i represents a position of each of the sample points and a sequence of the sample points in the wavelet signal sequence, and imod represents a position and a sequence of the ith sample point in the preset duration. When the preset duration arrives, imod may be reset and restart to record a position of each of sample points of a next voice signal in the next preset duration.
At 1208, whether imod is equal to Vmin is determined. Specifically, imod and Vmin are compared to determine whether tracking for the sample points has reached the preset duration. In one example, if perform three-stage wavelet packet decomposition and down-sampling with 16 KHz as a sampling frequency of the first audio signal, sampling is performed every 0.5 ms in the wavelet signal sequence, the accumulation sample point number Vmin=10, and thus a tracking duration is Vwin×0.5=5 ms. If imod is equal to Vmin, it means that the preset tracking duration is reached, and proceed to operations at 1209. Otherwise, if imod is not equal to Vmin, optionally, if imod is less than Vmin, proceed to operations at 1213.
At 1209, imod=0. Specifically, each time imod reaches the accumulation sample point number Vmin, imod is released and is then reset (e. g., imod=0) to perform a next sample point accumulation.
At 1210, whether i is equal to Vmin is determined. Specifically, when i is equal to Vmin, proceed to operations at 1211 to initialize matrix data. When i is not equal to Vmin, proceed to operations at 1212.
At 1211, a matrix SW is initialized. Specifically, define SW:
When i is equal to Vmin, define a matrix SW with Nwin rows and one column. Optionally, Nwin=2. It can be understood that the operation is performed at the beginning of a voice, i is increased all the time, and Vmin is a preset fixed value. When the (Vmin)th sample point is traversed by i, the matrix SW is initialized, to provide a matrix to store data in embodiments.
At 1212, data in the matrix SW is updated, a minimum value in the matrix is recorded by Smin=min{SW}, and Smact is reset (e.g., Smact=S(i)). Specifically, SW is:
When i is not equal to Vmin, and accumulation index imod indicates that the preset duration is reached, values in the matrix SW is updated. A minimum value among audio intensity values of all the sample points in a present time period and a minimum value in a previous time period are stored in the matrix SW, and then the smaller of the two values is obtained and recorded by Smin, i.e., Smin=min {SW}. It can be understood that Smin records a minimum value among audio intensity values of all sample points starting from a sample point previous to a (Vmin)th sample point. Smact is released and is then reset, e.g., Smact=S(i). In one example, a tracking duration of 5 ms is taken as an example for illustration, Smact records a minimum value among fourth reference audio intensity values of all sample points in a latest 5 ms, and Smin records a minimum value among fourth reference audio intensity values of all sample points in a previous 5 ms. Thereafter, minimum values in two adjacent 5 ms are stored in the matrix SW of length 2, and then the smaller of the two values is obtained and is recorded by Smin, i.e., Smin=min {SW}. Smin that records the minimum value in the tracking duration is assigned to Sm(i) at 1206, i.e., Sm(i)=Smin.
At 1213, whether i is greater than or equal to the total number of sample points is determined. Specifically, it can be understood that after operations at 1202 (i=i+1) is performed, before another signal of the preset time period is tracked, it is necessary to determine a position of the sample point in the wavelet signal sequence, i.e., determine whether i of the ith sample point is greater than or equal to the total number of sample points in the wavelet signal sequence because i is kept increasing by one to sequentially traverse subsequent sample points. If i is less than the total number of sample points in the wavelet signal sequence, proceed to signal tracking. If the last of all the sample points is traversed by i, e.g., i is equal to or greater than the total number of sample points, proceed to operations at 1214.
At 1214, Sm(i) is determined as a first reference audio intensity value or audio intensity value of the ith sample point. Specifically, it can be obtained from steps 1212 and 1206 that Sm(i) records the minimum value among audio intensity values of all sample points starting from a sample point previous to a (Vmin)th sample point. In some possible implementations, Sm(i) is the first reference audio intensity value of the ith sample point, such that implementations described at 1003 of embodiments described above in conjunction with
In embodiments, the minimum value Smin among the audio intensity values of all the sample points in a previous tracking duration is passed to the present tracking duration by means of the matrix, and then Smin is compared with the audio intensity value of the target sample point. The minimum value among the fourth reference audio intensity values of the sample points including the target sample point and all the sample points previous to the target sample point in the wavelet signal sequence is obtained, that is, Smin=min (Smin, S(i)), which is then determined as the first reference audio intensity value Sm(i) of the target sample point. Thereafter, the smaller of the two values (i.e., Smin and the audio intensity value of the target sample point) is compared with a fourth reference audio intensity value of a sample point subsequent to the target sample point to obtain a smallest value among the three values, and then the smallest value is determined as a first reference audio intensity value Sm(i+1) of a sample point subsequent to the target sample point. In above manners, the minimum value among the audio intensity values of all the sample points in the tracking duration is obtained, and a smaller value between the minimum value among audio intensity values in the previous tracking duration and a minimum value among audio intensity values in the present tracking duration is passed to a next tracking duration by means of the matrix. The sample point sequence formed by Sm(i) can describe the distribution of the audio intensity values of the voice signal, which can also be understood as the energy distribution trend of the voice signal.
By implementing embodiments, by tracking the audio intensity values of the signal of the stable duration, the accuracy of detection of the valid voice signal can be further improved, such that the false detection that transient noise is determined as a valid voice signal or valid voice signal segment can be further avoided.
The effect of implementing embodiments can be exemplarily described below with reference to the accompanying drawings. Referring to
After the device for detecting the valid signal performs the wavelet decomposition or wavelet packet decomposition described above in conjunction with
To further reduce the influence of signal burr, the original signal amplitude and the audio intensity values of all the sample points after the steady-state amplitude tracking are smoothed, and the smoothed result is illustrated in
The detection of the valid voice signal, that is, voice activity detection (VAD), is performed on the signal in
The following describes the device for detecting a valid signal provided in the embodiments of the disclosure. Referring to
The obtaining module 1401 is configured to obtain a first audio signal of a preset duration, where the first audio signal includes at least one audio frame signal.
The decomposition module 1402 is configured to obtain multiple wavelet decomposition signals respectively corresponding to the at least one audio frame signal by performing wavelet decomposition on each audio frame signal, where each wavelet decomposition signal contains multiple sample points and an audio intensity value of each sample point.
The combining module 1403 is configured to obtain a wavelet signal sequence by combining the wavelet decomposition signals corresponding to the at least one audio frame signal according to a framing sequence of the at least one audio frame signal in the first audio signal.
The determining module 1404 is configured to obtain a maximum value and a minimum value among audio intensity values of all sample points in the wavelet signal sequence, and determine a first audio intensity threshold according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence.
The determining module 1404 is further configured to obtain sample points each having an audio intensity value greater than the first audio intensity threshold in the wavelet signal sequence, and determine a signal of sample points in the first audio signal corresponding to the sample points each having an audio intensity value greater than the first audio intensity threshold in the wavelet signal sequence as the valid voice signal.
In some possible embodiments, the determining module 1404 is further configured to determine the first audio intensity threshold and a second audio intensity threshold according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, where the first audio intensity threshold is less than the second audio intensity threshold.
The obtaining module 1401 is further configured to obtain a first sample point in the wavelet signal sequence, where an audio intensity value of a sample point previous to the first sample point is less than the second audio intensity threshold, and an audio intensity value of the first sample point is greater than the second audio intensity threshold.
The obtaining module 1401 is further configured to obtain a second sample point in the wavelet signal sequence, where the second sample point is after the first sample point and is the first of sample points each having an audio intensity value less than the first audio intensity threshold in the wavelet signal sequence.
The determining module 1404 is further configured to determine a signal of the sample points in the first audio signal corresponding to sample points from the first sample point to a sample point previous to the second sample point in the wavelet signal sequence as a valid voice segment in the valid voice signal.
Optionally, at least a first preset number of consecutive sample points are included between the second sample point and the first sample point.
In some possible embodiments, the determining module 1404 is further configured to determine an average value of first reference audio intensity values of a second preset number of consecutive sample points including a target sample point in the wavelet signal sequence as an audio intensity value of the target sample point.
In some possible implementations, the device 14 for detecting a valid voice signal further includes a calculating module 1405. Before the determining module 1404 determines the average value of the first reference audio intensity values of the second preset number of consecutive sample points including the target sample point in the wavelet signal sequence as the audio intensity value of the target sample point, the calculating module 1405 is configured to obtain a second reference audio intensity value of the target sample point by multiplying an audio intensity value of a sample point previous to the target sample point in the wavelet signal sequence by a smoothing coefficient. The calculating module 1405 is further configured to obtain a third reference audio intensity value of the target sample point by multiplying an average value of audio intensity values of sample points that include the target sample point and all consecutive sample points previous to the target sample point in the wavelet signal sequence by a remaining smoothing coefficient. The calculating module 1405 is further configured to determine a sum of the second reference audio intensity value and the third reference audio intensity value as a fourth reference audio intensity value of the target sample point. The determining module 1404 is further configured to determine a minimum value among fourth reference audio intensity values of sample points including the target sample point and all sample points previous to the target sample point in the wavelet signal sequence as the first reference audio intensity value of the target sample point.
Optionally, the determining module 1404 is further configured to determine a value obtained by processing a reference maximum value of each of all the wavelet decomposition signals in the wavelet signal sequence as the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, and determine a value obtained by processing a reference minimum value of each of all the wavelet decomposition signals in the wavelet signal sequence as the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, where for each wavelet decomposition signal, the reference maximum value of the wavelet decomposition signal is obtained according to a maximum value among audio intensity values of all sample points of the wavelet decomposition signal, and the reference minimum value of the wavelet decomposition signal is obtained according to a minimum value among audio intensity values of all sample points of the wavelet decomposition signal.
In some possible embodiments, the device 14 for detecting a valid voice signal further includes a compensating module 1406. Before the obtaining module 1401 obtains the first audio signal of the preset duration, the compensating module 1406 is further configured to obtain the first audio signal by compensating for a high-frequency component in an original audio signal of the preset duration.
In some possible implementations, the decomposition module 1402 is further configured to perform wavelet packet decomposition on each audio frame signal, and determine each signal obtained after the wavelet packet decomposition as the wavelet decomposition signal.
In some possible implementations, the determining module 1404 is further configured to determine a first audio intensity threshold according to TL=min(λ1·(Scmax−Scmin)+Scmin, λ2·Scmin) and the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence. Scmax represents the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, Scmin represents the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, λ1 represents a second preset threshold, and λ2 represents a third preset threshold.
In other possible implementations, the determining module 1404 is further configured to determine a first audio intensity threshold according to TL=min(λ1·(Scmax−Scmin)+Scmin, λ2·Scmin) and the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, wherein Scmax represents the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, Scmin represents the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, λ1 represents a second preset threshold, and λ2 represents a third preset threshold; and determine the second audio intensity threshold according to TU=αTL, where α represents a fourth preset threshold and is greater than 1.
It can be understood that, in embodiments, for the specific implementation of detection of the valid voice signal, reference may be made to embodiments described above in conjunction with
According to embodiments, by collecting the energy information of all the sample points in the wavelet signal sequence, according to the energy distribution of all the sample points in the wavelet signal sequence, determination and detection of the valid voice signal may be achieved, such that the accuracy of detection of the valid voice can be improved.
The following describes an apparatus for detecting a valid signal provided in the embodiments of the disclosure. Referring to
The transceiver 1500 is coupled with the processor 1501 and the memory 1502, and the processor 1501 is further coupled with the memory 1502.
The transceiver 1500 is configured to obtain a first audio signal of a preset duration, where the first audio signal includes at least one audio frame signal.
The processor 1501 is configured to obtain multiple wavelet decomposition signals respectively corresponding to the at least one audio frame signal by performing wavelet decomposition on each audio frame signal, where each wavelet decomposition signal contains multiple sample points and an audio intensity value of each sample point.
The processor 1501 is further configured to obtain a wavelet signal sequence by combining the wavelet decomposition signals corresponding to the at least one audio frame signal according to a framing sequence of the at least one audio frame signal in the first audio signal.
The processor 1501 is further configured to obtain a maximum value and a minimum value among audio intensity values of all sample points in the wavelet signal sequence, and determine a first audio intensity threshold according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence.
The processor 1501 is further configured to obtain sample points each having an audio intensity value greater than the first audio intensity threshold in the wavelet signal sequence, and determine a signal of sample points in the first audio signal corresponding to the sample points each having an audio intensity value greater than the first audio intensity threshold in the wavelet signal sequence as the valid voice signal.
The memory 1502 is configured to store computer programs, and the computer programs are invoked by the processor 1501.
In some possible embodiments, the processor 1501 is further configured to: determine the first audio intensity threshold and a second audio intensity threshold according to the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, where the first audio intensity threshold is less than the second audio intensity threshold; obtain a first sample point in the wavelet signal sequence, where an audio intensity value of a sample point previous to the first sample point is less than the second audio intensity threshold, and an audio intensity value of the first sample point is greater than the second audio intensity threshold; obtain a second sample point in the wavelet signal sequence, where the second sample point is after the first sample point and is the first of sample points each having an audio intensity value less than the first audio intensity threshold in the wavelet signal sequence; and determine a signal of the sample points in the first audio signal corresponding to sample points from the first sample point to a sample point previous to the second sample point in the wavelet signal sequence as a valid voice segment in the valid voice signal.
Optionally, a first preset number of consecutive sample points are included between the second sample point and the first sample point.
In some possible embodiments, the processor 1501 is further configured to determine an average value of first reference audio intensity values of a second preset number of consecutive sample points including a target sample point in the wavelet signal sequence as an audio intensity value of the target sample point.
In some possible embodiments, the processor 1501 is further configured to: obtain a second reference audio intensity value of the target sample point by multiplying an audio intensity value of a sample point previous to the target sample point in the wavelet signal sequence by a smoothing coefficient; obtain a third reference audio intensity value of the target sample point by multiplying an average value of audio intensity values of sample points that include the target sample point and all consecutive sample points previous to the target sample point in the wavelet signal sequence by a remaining smoothing coefficient; determine a sum of the second reference audio intensity value and the third reference audio intensity value as a fourth reference audio intensity value of the target sample point; and determine a minimum value among fourth reference audio intensity values of sample points including the target sample point and all sample points previous to the target sample point in the wavelet signal sequence as the first reference audio intensity value of the target sample point.
In some possible implementations, the processor 1501 is further configured to determine a value obtained by processing a reference maximum value of each of all the wavelet decomposition signals in the wavelet signal sequence as the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, and determine a value obtained by processing a reference minimum value of each of all the wavelet decomposition signals in the wavelet signal sequence as the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, where for each wavelet decomposition signal, the reference maximum value of the wavelet decomposition signal is obtained according to a maximum value among audio intensity values of all sample points of the wavelet decomposition signal, and the reference minimum value of the wavelet decomposition signal is obtained according to a minimum value among audio intensity values of all sample points of the wavelet decomposition signal.
Optionally, the processor 1501 is further configured to: obtain the first audio signal by compensating for a high-frequency component in an original audio signal of the preset duration.
In some possible embodiments, the processor 1501 is further configured to: perform wavelet packet decomposition on each audio frame signal, and determine each signal obtained after the wavelet packet decomposition as the wavelet decomposition signal.
In some possible implementations, the processor 1501 is further configured to: determine a first audio intensity threshold according to TL=min(λ1·(Scmax−Scmin)+Scmin, λ2·Scmin) and the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence. Scmax represents the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, Scmin represents the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, λ1 represents a second preset threshold, and λ2 represents a third preset threshold.
In some possible implementations, the processor 1501 is further configured to: determine a first audio intensity threshold according to TL=(Scmax−Scmin)+Scmin, λ2·Scmin) and the maximum value and the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, where Scmax represents the maximum value among the audio intensity values of all the sample points in the wavelet signal sequence, Scmin represents the minimum value among the audio intensity values of all the sample points in the wavelet signal sequence, λ1 represents a second preset threshold, and λ2 represents a third preset threshold; and determine the second audio intensity threshold according to TU=αTL, where α represents a fourth preset threshold and is greater than 1.
It can be understood that the apparatus 15 for detecting a valid signal can perform implementations provided in the operations in the above-mentioned
According to embodiments, when the apparatus for detecting a valid voice signal detects a valid voice signal, other working modules of the apparatus can be woken up, thereby reducing power consumption of the apparatus.
A readable storage medium is further provided in the disclosure. The readable storage medium stores instructions, and the instructions are executed by a processor of the apparatus for detecting a valid voice signal to implement operations of the method in the various aspects of
It should be noted that the above-mentioned terms “first” and “second” are merely for illustration, and should not be construed as indicating or implying relative importance.
In embodiments of the disclosure, the energy information of all the sample points in the wavelet signal sequence can be collected, and determination and detection of the valid voice signal may be achieved according to the energy distribution of the wavelet signal sequence, which can improve the accuracy of detection of the valid voice. In addition, the audio intensity values of all the sample points in the wavelet signal sequence are smoothed and the energy distribution information of all the sample points in the wavelet signal sequence may be tracked, such that the accuracy of the detection of the valid voice signal can be further improved.
In several embodiments provided in the disclosure, it can be understood that the method, device/apparatus, and system disclosed in embodiments provided herein may be implemented in other manners. For example, the embodiments described above are merely illustrative; for instance, the division of the unit is only a logical function division and there can be other manners of division during actual implementations, for example, multiple units or components may be combined or may be integrated into another system, or some features may be ignored or not performed. In addition, coupling, or direct coupling, or communication connection between each illustrated or discussed components may be indirect coupling or communication connection among devices or units via some interfaces, devices, and units, and may be electrical connection, mechanical connection, or other forms of connection.
The units described as separate components may or may not be physically separated, the components illustrated as units may or may not be physical units, that is, they may be in the same place or may be distributed to multiple network units. All or part of the units may be selected according to actual needs to achieve the purpose of the technical solutions of the embodiments.
In addition, the functional units in various embodiments of the disclosure may be integrated into one processing unit, or each unit may be physically present, or two or more units may be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or can be implemented in the form of hardware in combination with a software function unit.
It will be understood by those of ordinary skill in the art that all or a part of the various methods of the embodiments described above may be accomplished by means of a program to instruct associated hardware, the program may be stored in a computer-readable storage medium. The program, when executed, can implement operations including the method embodiments. The storage medium may include: a removable storage device, read-only memory (ROM), random access memory (RAM), a magnetic disk, an optical disk, or other media that can store program codes.
Alternatively, the integrated unit may be stored in a computer-readable storage medium when it is implemented in the form of a software functional module and is sold or used as a separate product. Based on such understanding, the technical solutions of the disclosure essentially, or the part of the technical solutions that contributes to the related art may be embodied in the form of a software product. The computer software product is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the method described in the various embodiments of the disclosure. The storage medium includes various medium capable of storing program codes, such as a removable storage device, a ROM, a RAM, a magnetic disk, an optical disk, or the like.
The above are merely specific embodiments of the disclosure, but the protection scope of the disclosure is not limited to this. Various changes or substitutions made by any person skilled in the art within the technical scope disclosed in the disclosure should be covered within the protection scope of the disclosure. Therefore, the protection scope of the disclosure should be subject to the protection scope of the claims.
| Number | Date | Country | Kind |
|---|---|---|---|
| 201911109218.X | Nov 2019 | CN | national |
The application is a continuation under 35 U.S.C. § 120 of International Application No. PCT/CN2020/128374, filed on Nov. 12, 2020, which claims priority under 35 U.S.C. § 119(a) and/or PCT Article 8 to Chinese Patent Application No. 201911109218.X, filed on Nov. 13, 2019, the entire disclosures of which are hereby incorporated by reference.
| Number | Date | Country | |
|---|---|---|---|
| Parent | PCT/CN2020/128374 | Nov 2020 | US |
| Child | 17728198 | US |
