Patent Application
20130077802
The present invention relates to a signal processing technology for suppressing noise included in a deteriorated signal and thus enhancing a desired signal.
A noise suppressing technology is known as a signal processing technology for suppressing a part or the whole of noise included in a deteriorated signal (a signal in which a desired signal and noise are mixed) and thus outputting an enhanced signal (a signal in which the desired signal is enhanced). For example, a noise suppressor is a system for suppressing noise superposed on a desired voice signal, which is used in a variety of voice terminals such as a mobile phone.
Related to this kind of technology, patent document 1 discloses a method which suppresses noise by multiplying an input signal by a suppression factor smaller than 1. Patent document 2 discloses a method which suppresses noise by directly subtracting estimated noise from a deteriorated signal.
In the technologies described in patent documents 1 and 2, it is necessary to estimate noise from a desired signal which is already mixed with the noise and thus deteriorated. However, there are limitations on accurate noise estimation from only a deteriorated signal. The methods described in patent documents 1 and 2 are effective, in general, only when noise is sufficiently small with respect to a desired signal. When the condition that noise is sufficiently small with respect to a desired signal is not satisfied, because accuracy of an estimated noise value is low, sufficient noise suppression effect is not obtained by the methods described in patent document 1 and 2, and in addition, large distortion is included in the enhanced signal.
On the other hand, patent document 3 discloses a technology which suppresses noise, when a characteristic of noise mixed into a desired signal can be known to some extent in advance, by subtracting noise information (information about a characteristic of noise) recorded in advance from the deteriorated signal. It also discloses a method which multiplies the noise information by a large factor when an input signal power obtained by analyzing the input signal is high, and by a small factor when the input signal power is low, and then subtracts the multiplication result from the deteriorated signal.
[Patent Document 1] Japanese Patent Publication No. 4282227
[Patent Document 2] Japanese Patent Application Laid-Open No. 1996-221092
[Patent Document 3] Japanese Patent Application Laid-Open No. 2006-279185
However, there has been no technology which can suppress both of noise having unknown characteristics and noise having known characteristics. For example, it has been impossible, when one performs verbal communication while manipulating a mobile phone, one performs video recording during an interview, and the like, to enhance a human speaking voice while suppressing mechanical noise and background noise.
In view of the above situation, the objective of the present invention is to provide a signal processing technology which solves the problem described above.
In order to achieve the above-described objective, a device according to the present invention comprises a noise suppression means for suppressing noise included in an inputted deteriorated signal, wherein the noise suppression means includes a first output means which, for the purpose of suppressing known noise having known characteristics, outputs known-noise information using noise information stored in advance, and a second output means which, for the purpose of suppressing unknown noise having unknown characteristics, outputs unknown-noise information by estimating the unknown noise, and the noise suppression means performs noise suppression using the known-noise information and the unknown-noise information.
In order to achieve the above-described objective, a method according to the present invention outputs known-noise information using noise information stored in advance, for the purpose of suppressing known noise having known characteristics, and outputs unknown-noise information by estimating the unknown noise, for the purpose of suppressing unknown noise having unknown characteristics, and suppresses noise included in an inputted deteriorated signal using the known-noise information and the unknown-noise information.
In order to achieve the above-described objective, a program recording medium according to the present invention stores a signal processing program which causes a computer to execute: a first outputting step of outputting known-noise information using noise information stored in advance, for the purpose of suppressing known noise having known characteristics; a second outputting step of outputting unknown-noise information by estimating the unknown noise, for the purpose of suppressing unknown noise having unknown characteristics; and a noise suppressing step of suppressing noise included in an inputted deteriorated signal using the known-noise information and the unknown-noise information.
The present invention provides a signal processing technology which can suppress both of noise having unknown characteristics and noise having known characteristics.
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
Hereinafter, exemplary embodiments of the present invention will be described in detail as examples, with reference to the drawings. Here, constituents described in the following exemplary embodiments are shown only as examples, and it is not intended to limit technological scope of the present invention to only those.
As a first exemplary embodiment of an information processing device according to the present invention, description will be given of a device which suppresses a part or whole of noise included in a deteriorated signal (a signal in which a desired signal is mixed with noise), and thus outputs an enhanced signal (a signal in which the desired signal is enhanced).
In the present exemplary embodiment, assumed are the cases such as those where one takes a picture or manipulates a machine while making verbal communication on a mobile phone, where one manipulates a machine while making audio recording with an IC recorder, and where one manipulates a machine or a machine operates while making video recording of an interview. According to the present exemplary embodiment, it is possible to achieve, in a well-balanced manner, both of suppression of noise generated around a device (background noise existing in surroundings of a user of the device or a speaker) and suppression of noise generated in the devise itself (known noise such as auto-focusing noise, zooming noise and shutter noise).
As such information processing devices, mentioned are digital still cameras, digital camcorders, laptop PCs, mobile phones, IC recorders and TV conference systems, for example. However, the present invention is not limited to them, but can be applied to any devices for which noise elimination is demanded.
The known-noise information output section 301 outputs known-noise information using noise information stored in advance in a storage section 311, in order to suppress known noise having known characteristics. The storage section 311 includes a storage device such as a semiconductor memory and stores information (noise information) about a characteristic of known noise as a suppression target. The noise stored as a suppression target is, for example, a shutter noise, motor driving noise, zooming noise, focusing noise of an auto-focusing mechanism (clicking sound) or the like. On the other hand, in order to suppress unknown noise having unknown characteristics, the unknown-noise information output section 303 includes an estimation section 331 for estimating unknown noise included in a deteriorated signal.
With the above configuration, the information processing device of the present exemplary embodiment can suppress both of noise having unknown characteristics and noise having known characteristics.
Here, the noise suppression unit 23 of the present exemplary embodiment performs further noise suppression, using unknown noise information outputted from the unknown-noise information output section 233, on a signal having undergone noise suppression using known noise information outputted from a known-noise information output section 231. That is, known noise information outputted from the known-noise information output section 231 is supplied to the known-noise suppression section 232, where it is used for suppressing known noise included in an inputted deteriorated signal. Further, a signal in which the known noise has been already suppressed is supplied to the unknown-noise information output section 233, where it is used for estimation of unknown noise by the unknown-noise information output section 233. Then, unknown-noise information outputted as a result of the estimation is outputted to an unknown-noise suppression section 234, where processing for suppressing the estimated unknown noise is performed on the signal in which the known noise has been already suppressed, and the resulting signal is outputted as an enhanced signal.
With the above configuration, the information processing device of the present exemplary embodiment achieves suppression of both noise having an unknown characteristic and that having a known characteristic.
In the present exemplary embodiment, the unknown-noise information output section 333 estimates unknown noise using an inputted deteriorated signal. Then, unknown-noise information outputted as a result of the estimation is outputted to the unknown-noise suppression section 234, where processing for suppressing the estimated unknown noise is performed on the signal in which known noise has been already suppressed, and the resulting signal is outputted as an enhanced signal.
With the above configuration, highly accurate estimation of unknown noise becomes possible even when accuracy of noise suppression by the known-noise information output section 231 and the known-noise suppression section 232 is low.
In the present exemplary embodiment, the known-noise information output section 431 generates known-noise information from noise information stored in a storage section, using an inputted deteriorated signal. For example, the known-noise information output section 431 analyzes an inputted deteriorated signal and, by a mixing method according to the analysis result, mixes pieces of noise information stored in advance together, and thereby generates and outputs mixed noise information (pseudo noise information) as known-noise information. At least one of a plurality of pieces of noise information subjected to the mixing is one stored in the storage section in advance.
A more specific example is such that a plurality of pieces of noise information are stored in the storage section in advance, and the known-noise information output section 431 mixes them into a combination. Targets of the mixing are, for example, with respect to known noise, a combination of maximum and average noise information, a combination of maximum, average and minimum noise information, a combination of peak component noise information and others, a combination of impact component noise information and others, and the like.
Here definitions of the average, maximum, minimum, peak component and impact component noise information are as follows, respectively.
Average noise information: one obtained, with respect to whole (a plurality of frames) of known noise, by averaging amplitudes (or powers) of the same frequency component in a plurality of spectra derived by Fourier transform; which is what is called an average spectrum obtained by averaging in terms of time.
Maximum noise information: a maximum value of amplitude (or power) for each frequency component in a plurality of spectra derived by Fourier transform of the whole (a plurality of frames) of known noise; which is a so-called maximum spectrum.
Minimum noise information: a minimum value of amplitude (or power) for each frequency component in a plurality of spectra derived by Fourier transform of the whole (a plurality of frames) of known noise; which is a so-called minimum spectrum.
Peak component noise information: a frequency component having a distinctly large amplitude value compared to that of neighboring components, when the amplitudes are compared with each other in sequence of frequency in spectra derived by Fourier transform of the whole (a plurality of frames) of known noise.
Impact component noise information: an average of a plurality of spectra derived by Fourier transform of the whole of impact sounds; which is a so-called average spectrum of impact sounds. Although an impact sound is itself one having a large amplitude value in an extremely short period of time when change with time of its audio signal before Fourier transform is observed, its spectrum after Fourier transform is characterized by that the amplitude is almost constant over a certain frequency range.
As described above, by mixing pieces of noise information stored in advance by the use of a mixing method according to a result of analysis on an inputted deteriorated signal, and thus generating mixed noise information (pseudo noise information) as known-noise information, it is possible to realize noise suppression capable of dealing with a heavily changing signal characteristic.
In the noise suppression unit 53, the selection section 535 calculates, for example, a gain g1 (0<g1<1) with respect to known-noise information and a gain g2 (0<g2<1) with respect to unknown-noise information, selects either the smaller or the larger one between them and supplies it to a suppression section 536. The suppression section 536 performs noise suppression by multiplying a deteriorated signal by the gain supplied from the selection section 535. Here, the noise suppression by the noise suppression unit 53 is performed for each frequency component of the deteriorated signal. Specifically, for each frequency, the noise suppression unit 53 evaluates which is more serious between known noise and unknown noise, and performs noise suppression in a way adapted to the more serious one.
Here, the configuration may be such that a deteriorated signal is inputted to the selection section 535, and the selection section 535 determines, by analyzing the deteriorated signal, which of known-noise information and unknown-noise information to use for the suppression. In this case, the suppression section 536 performs noise suppression by subtracting the known-noise or unknown-noise information supplied from the selection section 535 from the deteriorated signal.
According to the present exemplary embodiment, noise suppression can be realized more efficiently. For example, if the selection section 535 selects a minimum gain value, more serious noise can be suppressed effectively. That is, according to the present exemplary embodiment, it is possible, in a situation where known noise is overall larger than unknown noise, to preferentially suppress known noise at a frequency where known noise is larger, and unknown noise at a frequency where known noise is smaller. On the other hand, if the selection section 535 selects a maximum gain value, occurrence of distortion due to noise suppression can be prevented effectively, and high sound quality can thus be realized. Further, when a threshold gain value Gth for no occurrence of distortion is known, the noise suppression unit 53 may perform the suppression using a smaller one of the gains g1 and g2 if both of the gains are larger than the threshold value Gth, and using the threshold value Gth in place of the smaller one if either of the gains is smaller than the threshold value Gth.
As in the present exemplary embodiment, by generating known-noise information according to an analysis result on an inputted deteriorated signal, noise suppression capable of dealing with a heavily changing signal characteristic can be achieved, in addition to the effect of the fifth exemplary embodiment.
In the noise suppression unit 73, a mixing section 735 calculates, for example, a gain g1 (0<g1<1) with respect to known-noise information and a gain g2 (0<g2<1) with respect to unknown-noise information, and supplies a value obtained by mixing them (a medium value, for example) to a suppression section 536. The suppression section 536 performs noise suppression by multiplying a deteriorated signal by the gain supplied from the mixing part 735. Here, the noise suppression by the noise suppression unit 73 is performed for each frequency component of the deteriorated signal. That is, known noise and unknown noise are suppressed for each frequency.
Here, the configuration may be such that a deteriorated signal is inputted to the mixing section 735, and the mixing section 735 analyzes the deteriorated signal and mixes known-noise information and unknown-noise information according to the analysis result, and the mixed information is used for the suppression. That is, the mixing section 735 may mix known-noise information and unknown-noise information by weighting them according to the analysis result on the inputted deteriorated signal.
According to the present exemplary embodiment, it is possible to obtain more accurately a gain of suppression or a component to subtract for suppression, and thus to realize noise suppression effectively. As a result, it becomes possible to realize noise suppression and prevention of distortion occurrence due to it in a well-balanced manner.
As in the present exemplary embodiment, by generating known-noise information according to an analysis result on an inputted deteriorated signal, noise suppression capable of dealing with a heavily changing signal characteristic can be achieved, in addition to the effect of the seventh exemplary embodiment.
In the present exemplary embodiment, using
As shown in
A deteriorated signal is supplied to the input port 1 as a series of sample values. The deteriorated signal supplied to the input port 1 undergoes transformation such as Fourier transform at the transformation unit 2, and thus is divided into a plurality of frequency components. An amplitude spectrum of the plurality of frequency components is supplied to the noise suppression unit 3, and a phase spectrum is transmitted to the inverse transformation unit 4. Although an amplitude spectrum is supplied to the noise suppression unit 3 here, the present invention is not limited to it, but a power spectrum corresponding to the square of the amplitude spectrum may be supplied to the noise suppression unit 3.
The noise suppression unit 3 suppresses noise at each frequency of the deteriorated-signal amplitude spectrum supplied from the transformation unit 2, and transmits to the inverse transformation unit 4 an enhanced-signal amplitude spectrum as a result of the noise suppression. The inverse transformation unit 4 performs inverse transformation combining the enhanced-signal amplitude spectrum supplied from the noise suppression unit 3 and the phase spectrum of the deteriorated signal supplied from the transformation unit 2, and supplies the result as an enhanced signal sample to the output port 5
[Configuration of the Transformation Unit]
n(t)=w(t)yn(t) (1)
Windowing with an overlap of a part of successive two frames also is widely used. Assuming an overlap length to be 50% of the frame length, for t=0, 1, . . . , K/2−1, the left-hand sides of the following equation (2) become the output of the windowing section 122.
A bilaterally-symmetric window function is used for a real numbered signal. A window function is designed in a manner where, when a suppression factor is set to be 1 in the MMSE STSA (Minimum Mean-Square Error Short-Time Spectral Amplitude Estimator) method, or when zero is subtracted in the SS (Spectrum Subtraction) method, an input signal and an output signal coincide with each other except for a calculational error. This means that w(t)+w(t+K/2)=1 holds.
Hereinafter, description will be continued taking as an example the case where windowing is performed with 50% overlap between successive two frames. As w(t), for example, Hanning window expressed by the following equation (3) can be used.
Besides it, also known are a variety of window functions such as Hamming,
Kaiser, and Blackman windows. The windowed output is supplied to the Fourier transform section 123, where it is transformed into a deteriorated signal spectrum Yn(k). The deteriorated signal spectrum Yn(k) is separated into an amplitude and a phase spectra, and the deteriorated-signal phase spectrum argYn(k) is supplied to the inverse transformation unit 4, and the deteriorated-signal amplitude spectrum |Yn(k)| to the noise suppression unit 3. As already described, a power spectrum can be used instead of an amplitude spectrum.
[Configuration of the Inverse Transformation Unit]
n(k)=|
The inverse Fourier transform section 143 performs inverse Fourier transform on the obtained enhanced signal. The enhanced signal having undergone inverse Fourier transform is supplied to the windowing section 142 in the form of a time-domain series of sample values xn(t) (t=0, 1, . . . , K−1) where each frame includes K samples, and there, it is multiplied by a window function w(t). A signal obtained by windowing an input signal xn(t) (t=0, 1, . . . , K/2−1) in the n-th frame with w(t) is given by the left-hand side of the following equation (5).
n(t)=w(t)xn(t) (5)
Windowing with overlap of a part of successive two frames also is widely used. Assuming an overlap length to be 50% of the frame length, for t=0, 1, . . . , K/2−1, the left-hand sides of the following equation (6) become the output of the windowing section 142, which is transmitted to the frame combining section 141.
The frame combining section 141 takes K/2 samples from each of successive two frames in the output from the windowing section 142, superposes them together, and calculates by the following equation (7) output signals for t=0, 1, . . . , K−1 (the left-hand side of the equation (7)). The obtained output signals are transmitted from the frame combining section 141 to the output port 5.
{circumflex over (x)}
n(t)=
Although the above description has been given assuming that the transform used at the transformation section 2 and the inverse transformation section 4 is Fourier transform in
Further, also possible is that the transformation unit 2 integrates a plurality of frequency components and then the noise suppression unit 3 performs actual suppression. In that case, by increasing the number of integrated frequency components with increasing frequency from a low frequency region where the resolution of the human auditory system is high toward a high frequency region where the resolution is low, the transformation unit 2 can achieve high sound quality. In this way, by performing noise suppression after integrating a plurality of frequency components, the number of frequency components to be subjected to noise suppression becomes smaller, and thus the whole calculation amount is reduced.
[Processing at the Noise Suppression Unit]
In the noise suppression unit 3, a variety of suppression can be performed. As representative suppression methods, there are the SS method and the MMSE STSA method. The SS method subtracts noise information from the deteriorated-signal amplitude spectrum supplied from the transformation unit 2. The MMSE STSA method calculates a suppression factor with respect to each of a plurality of frequency components, using noise information and the deteriorated-signal amplitude spectrum supplied from the transformation unit 2, and multiplies the deteriorated-signal amplitude spectrum by the calculated suppression factors. The suppression factors are determined in a manner to minimize the mean-square power of an enhanced signal.
In noise suppression, the noise suppression unit 3 may adopt flooring so as to avoid excessive suppression. The flooring is a method for avoiding suppression exceeding a maximum suppression amount, and a flooring parameter determines the maximum suppression amount. The SS method imposes a restriction so that a result of subtracting corrected noise information from the deteriorated-signal amplitude spectrum does not become smaller than a flooring parameter. Specifically, when the subtraction result is smaller than the flooring parameter, the SS method substitutes the subtraction result with the flooring parameter. The MMSE STSA method substitutes a suppression factor with a flooring parameter when the suppression factor calculated from corrected noise information and the deteriorated-signal amplitude spectrum is smaller than the flooring parameter. Details of the flooring are disclosed in a document: M. Berouti, R. Schwartz and J. Makhoul, “Enhancement of speech corrupted by acoustic noise,” Proceedings of ICASSP '79, pp. 208-211, April 1979. By employing a flooring parameter, the noise suppression unit 3 suffers no occurrence of excessive suppression and can prevent increase in distortion of an enhanced signal.
The noise suppression unit 3 may also set the number of frequency components of noise information to be smaller than that of a deteriorated-signal spectrum. In that case, a plurality of pieces of noise information are commonly used for a plurality of frequency components. Because frequency resolution of a deteriorated-signal spectrum is higher in this case than in the case where a plurality of frequency components are integrated with respect to both a deteriorated-signal spectrum and noise information, the noise suppression unit 3 can achieve high sound quality with a smaller calculation amount than when no integration of frequency components is performed. Details of the suppression using noise information with a smaller number of frequency components than that of a deteriorated-signal spectrum are disclosed in Japanese Patent Application Laid-Open No. 2008-203879.
A tenth exemplary embodiment of the present invention will be described below using
In the present exemplary embodiment, by multiplying noise information (for example, a gain) on noise to suppress by a scaling factor based on an enhanced-signal amplitude spectrum as a noise suppression result, the noise suppression unit 3 corrects the noise information.
When updating the scaling factor using a fed-back noise suppression result, the noise suppression unit 3 performs updating in a manner such that the larger the noise suppression result with no desired signal in the deteriorated signal is (the larger the residual noise is), the larger the noise information after the correction becomes. It is because a large noise suppression result with no desired signal indicates that the suppression is insufficient, and it is desirable to increase noise information after the correction by changing the scaling factor. If noise information after the correction is large, a value to be subtracted becomes large in the SS method, and accordingly the noise suppression result becomes small. In multiplication type suppression such as the MMSE STSA method, an estimated value of the signal to noise ratio used for calculating a suppression factor becomes low, and accordingly a small suppression factor is obtained. These bring about stronger noise suppression. For updating a scaling factor, a plurality of methods can be considered. As examples, a recalculation method and a successive updating method will be described below.
As a noise suppression result, a state of complete noise suppression is ideal. For this purpose, the noise suppression unit 3 may recalculate or successively update the scaling factor in a manner, for example, to achieve a state of complete noise suppression when the amplitude or power of a deteriorated signal is small. It is because when the amplitude or power of a deteriorated signal is small, it is more likely that the power of a signal other than the noise to be suppressed is also small. The noise suppression unit 3 can evaluate if the amplitude or power of a deteriorated signal is small by detecting that the amplitude or power of the deteriorated signal is smaller than a threshold value.
Also, the noise suppression unit 3 can evaluate if the amplitude or power of a deteriorated signal is small by detecting that a difference between the amplitude or power of the deteriorated signal and that of noise information stored in the storage section 311 is smaller than a threshold value. That is, the noise suppression unit 3 uses a fact that, when the amplitude or power of a deteriorated signal is comparable to that of noise information, occupancy of the noise information in the deteriorated signal is high (the signal-to-noise ratio is low). In particular, by using information at a plurality of frequency points in an integrated manner, it becomes possible for the noise suppression unit 3 to compare outlines of spectra and thus to increase detection accuracy.
A scaling factor in SS method is recalculated in a manner such that, for each frequency, corrected noise information becomes equal to a deteriorated-signal spectrum with no desired signal in the deteriorated signal. In other words, the noise suppression unit 3 calculates the scaling factor αn such that a deteriorated-signal spectrum |Yn(k)| supplied from the transformation unit 2 when there is only noise coincides with a product of the scaling factor αn and noise information ν(k). Here, n is a frame index, and k is a frequency index. Accordingly, the scaling factor αn(k) is calculated by the following equation (8).
αn(k)=|Yn(k)|/νn(k) (8)
On the other hand, in successive updating of a scaling factor in SS method, a scaling factor is incrementally updated in a manner to make an enhanced-signal amplitude spectrum at a timing point with no desired signal inputted become closer to zero. When the noise suppression unit 3 uses the least mean square (LMS) algorithm for the successive updating, it calculates αn+1(k) by the following equation (9) using an error en(k) for the nth frame and a frequency index k.
αn+1(k)=αn(k)+μen(k)/νn(k) (9)
Here, μ is a small constant referred to as a step size.
When the noise suppression unit 3 immediately adopts the calculated scaling factor αn(k), it uses the following equation (10) in place of the equation (9).
αn(k)=αn−1(k)+μen(k)/νn(k) (10)
That is, the noise suppression unit 3 calculates a current scaling factor αn(k) using a current error, and adopts it immediately. By immediately updating the scaling factor, the noise suppression unit 3 can realize highly accurate noise suppression in real time.
When the noise suppression unit 3 uses the normalized least mean squares (NLMS) algorithm, it calculates a scaling factor αn+1(k) by the following equation (11) using an error en(k) described above.
αn+1(k)=αn(k)+μen(k)νn(k)/σn(k)2 (11)
The σn(k)2 is an average power of the noise information νn(k), and can be calculated using an average based on an FIR filter (a moving average using a sliding window), or that based on an IIR filter (leaky integration) or the like.
Also, the noise suppression unit 3 may calculate the scaling factor αn+1(k) by the following equation (12) using a perturbation method.
αn+1(k)=αn(k)+μen(k) (12)
Further, the noise suppression unit 3 may calculate the scaling factor αn+1(k) by the following equation (13) using a sign function sgn{en(k)} expressing only the sign of an error.
αn+1(k)=αn(k)+μ·sgn{en(k)} (13)
Similarly, the noise suppression unit 3 may use the least square (LS) algorithm and other adaptive algorithms. Further, the noise suppression unit 3 can immediately adopt an updated scaling factor, and for that purpose, it may update a scaling factor in real time by modifying the equations (11) to (13) with reference to the modification from the equation (9) to equation (10).
In the MMSE STSA method, a scaling factor is successively updated. The noise suppression unit 3 updates the scaling factor αn(k), for each frequency, by a method similar to that described above using the equations (8) to (13).
Comparing the recalculation method and the successive updating method described above as updating methods of a scaling factor, the former is characterized by a high tracking speed and the latter is by high accuracy. In order to take advantages of these characteristics, the noise suppression unit 3 can change an updating method such that the recalculation method is used first and the successive updating method is used later. In determining a timing to change an updating method, the noise suppression unit 3 may employ a condition if a scaling factor has become sufficiently close to an optimum value for changing the updating method. Alternatively, for example, the noise suppression unit 3 may change an updating method when a predetermined time has elapsed. Furthermore, the noise suppression unit 3 may also change an updating method when a correction amount of a scaling factor has become smaller than a predetermined threshold value.
As has been described above, according to the present exemplary embodiment, since noise information used for noise suppression is updated based on a noise suppression result, a variety of noise including unknown noise can be suppressed.
An eleventh exemplary embodiment of the present invention will be described below using
By this way, surefire noise suppression can be performed at a timing point when specific noise exists.
A twelfth exemplary embodiment of the present invention will be described below using
According to the present exemplary embodiment, since noise suppression is performed according to the proportion of the noise in the deteriorated signal, much more accurate noise suppression result can be obtained as a result.
Although, in the first to twelfth exemplary embodiments described above, descriptions have been given of information processing devices each having a different characteristic from the others, information processing devices obtained by combining those characteristics in any manner are also included within the scope of the present invention.
Further, the present invention may be applied to a system composed of a plurality of devises, and also may be applied to a sole device. Furthermore, the present invention can be applied to the case where a signal processing program of software for realizing the functions of the exemplary embodiments is supplied to a system or a device directly or from a remote place. Therefore, a program installed in a computer so as to realize the functions of the present invention, a medium storing the program and a WWW server allowing for downloading the program are included within the scope of the present invention.
By reading the signal processing program stored in the memory 1505, the CPU 1502 controls the whole operation of the computer 1500. That is, specifically, in order to suppress known noise having known characteristics, the CPU 1502 executing the signal processing program outputs known-noise information using noise information stored in advance (S1521). Next, in order to suppress unknown noise having unknown characteristics, the CPU 1502 estimates unknown noise and outputs unknown-noise information (S1522). Further, using the known-noise information and the unknown-noise information, the CPU 1502 suppresses the noise included in the deteriorated signal inputted at the input unit 1501 (S1523), and outputs an enhanced signal from the output unit 1504. By this way, the same effect as that of the first exemplary embodiment can be obtained.
A part or the whole of the above-described exemplary embodiments can be expressed also as the following further exemplary embodiments, but the present invention is not limited to them.
An information processing device comprising a noise suppression means for suppressing noise included in an inputted deteriorated signal, which is characterized by that
said noise suppression means comprises:
a first output means which outputs known-noise information, in order to suppress known noise having known characteristics, using noise information stored in advance; and
a second output means which estimates unknown noise, in order to suppress said unknown noise having unknown characteristics, and outputs unknown-noise information,
and said noise suppression means performs noise suppression using said known-noise information and said unknown-noise information.
The information processing device described in further exemplary embodiment 1, which is characterized by that,
on a signal having undergone noise suppression using the known-noise information outputted from said first output means, said noise suppression means further performs noise suppression using the unknown-noise information outputted from said second output means.
The information processing device described in further exemplary embodiments 1 or 2, which is characterized by that said second output means estimates said unknown noise, on the basis of said inputted deteriorated signal.
The information processing device described in further exemplary embodiments 1, 2 or 3, which is characterized by that
said first output means generates known-noise information from said stored noise information, on the basis of said inputted deteriorated signal.
The information processing device described in further exemplary embodiment 1, which is characterized by that
said noise suppression means further comprises
a selection means which selects either of said known-noise information outputted from said first output means or said unknown-noise information outputted from said second output means,
and performs noise suppression using noise information selected by said selection means.
The information processing device described in further exemplary embodiment 5, which is characterized by that
said selection means selects either of said known-noise information or said unknown-noise information, on the basis of said deteriorated signal.
The information processing device described in further exemplary embodiment 1, which is characterized by that
said noise suppression means
further comprises a mixing means for mixing said known-noise information outputted from said first output means and said unknown-noise information outputted from said second output means, and
performs noise suppression using mixed noise information generated in mixing by said mixing means.
The information processing device described in further exemplary embodiment 7, which is characterized by that said mixing means mixes said known-noise information and said unknown-noise information, on the basis of said deteriorated signal.
The information processing device described in any one of further exemplary embodiments 1 to 8, which is characterized by that it further comprises:
a transformation means for transforming an inputted voice signal into a spectrum and thus generating said deteriorated signal; and
an inverse transformation means for generating a voice signal to output by performing an inverse transform on a spectrum in which noise was suppressed by said noise suppression means.
The information processing device described in any one of further exemplary embodiments 1 to 9, which is characterized by that
said first output means outputs said known-noise information after correcting it according to a suppression result by said noise suppression means.
The information processing device described in any one of further exemplary embodiments 1 to 10, which is characterized by that
it further comprises a determination means for determining, in a deteriorated signal, how much a desired signal not to be suppressed exists or if it does not exist, and
said noise suppression means performs noise suppression according to the determination result by said determination means.
A signal processing method for suppressing noise included in an inputted deteriorated signal, which comprises:
outputting known-noise information using noise information stored in advance, in order to suppress known noise having known characteristics;
outputting unknown-noise information by estimating unknown noise having unknown characteristics, in order to suppress said unknown noise; and
suppressing noise included in said deteriorated signal using said known-noise information and said unknown-noise information.
A signal processing program for suppressing noise included in an inputted deteriorated signal, which is characterized that it makes a computer execute:
a first output step of outputting known-noise information using noise information stored in advance, in order to suppress known noise having known characteristics;
a second output step of outputting unknown noise information by estimating unknown noise having unknown characteristics, in order to suppress said unknown noise; and
a noise suppression step of suppressing noise included in said deteriorated signal using said known-noise information and said unknown-noise information.
Although the present invention has been described above with reference to the exemplary embodiments, the present invention is not limited to the above-described exemplary embodiments. In configurations and details of the present invention, various changes which are understood by those skilled in the art may be made within the scope of the present invention.
This application insists on priority based on Japanese Patent Application No. 2010-119495 proposed on May 25, 2010 and takes everything of the disclosure here.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2010-119495 | May 2010 | JP | national |
| Filing Document | Filing Date | Country | Kind | 371c Date |
|---|---|---|---|---|
| PCT/JP2011/061598 | 5/13/2011 | WO | 00 | 11/21/2012 |
