Patent Application
20160219381
This application claims the priority, under 35 U.S.C. §119, of German application DE 10 2015 201 073.2, filed Jan. 22, 2015; the prior application is herewith incorporated by reference in its entirety.
The invention relates to a method for suppressing noise in hearing aids and to a corresponding hearing aid. In this case, incoming audio signals are split into frequency bands.
Hearing aids are portable hearing apparatuses that are used for looking after people with impaired hearing. In order to meet the numerous individual needs, different designs of hearing aids are provided, such as behind the ear hearing aids (BTE), hearing aid with an external receiver (RIC: receiver in the canal) and in the ear hearing aids (ITE), e.g. including concha hearing aids or canal hearing aids (ITE, CIC). The hearing aids listed by way of example are worn on the outer ear or in the auditory canal. Furthermore, there are also bone conduction hearing aids, implantable or vibrotactile hearing aids available on the market, however. These involve the damaged hearing being stimulated either mechanically or electrically.
Hearing aids basically have the essential components of an input transducer, an amplifier and an output transducer. The input transducer is normally an acoustoelectric transducer, e.g. a microphone, and/or an electromagnetic receiver, e.g. an induction coil. The output transducer is generally in the form of an electroacoustic transducer, e.g. a miniature loudspeaker, or in the form of an electromechanical transducer, e.g. a bone conduction receiver. The amplifier is usually integrated in a signal processing device. The power supply is usually provided by a battery or a rechargeable storage battery.
For people with decreased hearing capacity, particularly speech comprehension is difficult with a high level of ambient noise, since the natural mechanisms for selection of particular sound sources in the signal processing in the brain are not effective, or are effective only to a decreased extent, owing to an input signal from the ear having decreased frequency range and dynamic range.
It is therefore necessary for the hearing aid to undertake some of these functions. In this case, it is already known practice for noise to be masked out by a directional characteristic, for example, or reduced using its spectral properties from an input signal from the microphones.
This also involves the use of adaptive filters that estimate or predict properties of a noise and filter a correspondingly estimated signal out of the input signal. Such methods are known, by way of example, from the publications by Rainer Martin, entitled “Noise Power Spectral Density Estimation Based on Optimal Smoothing and Minimum Statistics”, IEEE TRANSACTIONS ON SPEECH AND AUDIO PROCESSING, VOL. 9, NO. 5, July 2001, and by Timo Gerkmann and R. C. Hendriks, entitled “Unbiased MMSE-based Noise Power Estimation with Low Complexity and Low Tracking Delay”, IEEE TRANSACTIONS ON SPEECH AND AUDIO PROCESSING, VOL. 20, NO. 4, pages 1383-1393, May 2012. These documents describe how energy of a spurious signal can be estimated from an observation of temporal minima in an input signal.
However, in the case of speech, which contains a high dynamic range having fast changes, such noise suppression can result in undesirable artifacts that interfere with speech intelligibility.
It is therefore the object of the present invention to provide a hearing aid and a method for operating a hearing aid that improve speech comprehension given noise suppression.
The invention achieves this object by means of a method for operating a hearing aid and a hearing aid.
The method according to the invention for noise suppression in hearing aids has the step of splitting an audio signal into a plurality of essentially disjunct frequency bands.
In this case, the audio signal can originate from one or more microphones, a signal input or a wireless transmission device. The signal may be analog or digital in this case. The splitting into frequency bands can be effected using one or more filter means, for example a filter bank, a plurality of discrete filters or by virtue of transformation into a frequency space. In this case, essentially disjunct is intended to be understood to mean that the individual frequency bands overlap only to a small extent or not at all, for example by no more than one quarter, one tenth or one twentieth of their bandwidth.
One step of the method according to the invention involves a reference band from the plurality of frequency bands being selected that has an established first component of a speech signal.
In this case, it is conceivable for the hearing aid to have means for recognizing a speech component. The means can recognize the speech component on the basis of a spectral distribution, temporal dynamics, but also on the basis of a source direction for the audio signals picked up by a plurality of microphones in a hearing aid. A control section of the hearing aid can use this means to select a frequency band having a speech component as reference band.
One step of the method according to the invention involves the hearing aid ascertaining a correlation between a first frequency band and the reference band. This can be affected using a device for ascertaining a correlation, for example.
One step of the method according to the invention involves a value being ascertained on the basis of the ascertained correlation, which value indicates a second component of a speech signal in the first frequency band. This can be affected by the control section, for example. In this case, it is conceivable for a strong correlation between the first frequency band and the reference band to be able to be taken as a basis for inferring that the first frequency band also has a signal component containing speech. However, it is also conceivable for the first frequency band likewise to be rated with a means for recognizing a speech component. It is also possible for the means for recognizing a speech component to be applied only to the first frequency band when a sufficiently high correlation with the first frequency band has been ascertained.
A further step of the method according to the invention involves noise suppression in the first frequency band being set on the basis of the ascertained values.
The method according to the invention advantageously allows the noise suppression in the individual frequency bands to be made dependent on whether the frequency bands have a speech signal component. It is thus conceivable for no or just a reduced noise suppression to be applied when speech is present so as not to impair the intelligibility of the latter through artifacts.
Furthermore, the selection of a reference band allows recognition of the speech preferably in frequency ranges where this is simpler, e.g. owing to the speech spectrum, and transfer of the result also to other channels having a smaller component, in which channels this is more difficult, by using the correlation with the first frequency range to confirm the presence of a speech component.
In one conceivable embodiment, the method according to the invention having the steps of ascertainment of a correlation, ascertainment of a value and setting of noise suppression is carried out for a plurality of first frequency bands in parallel or sequentially.
It is thus advantageously possible to suppress a noise over a relatively large frequency range.
In one possible embodiment, the method according to the invention is repeated and, in so being, carried out with a second reference band. In this case, it is conceivable for the second reference band to be the same as the first reference band or to be different than the first reference band.
The use of different reference bands at different times allows the hearing aid to be matched to different situations with different speakers having different pitches and allows the speech to be reliably recognized under different conditions.
In one possible embodiment of the method according to the invention, a reference band is selected by selecting the frequency band having the highest energy in comparison with the other frequency bands from the plurality of frequency bands. In this case, the energy over the square of the amplitude is equivalent to the amplitude of the signal in the frequency band.
A single voice has, e.g. in the case of vowels, a high energy density in a narrow frequency range, which means that a voice component in a frequency band having high energy is probable.
In one conceivable embodiment of the method according to the invention, a reference band is selected by selecting the frequency band having a greatest degree of modulation in a predetermined frequency range.
A high degree of modulation in a frequency band indicates speech activity in this frequency band and can be ascertained with low processor load. A consideration of the degree of modulation in a frequency range that e.g. is characteristic of the modulation frequency of speech allows recognition certainty to be increased. In the method according to the invention, it is thus conceivable, by way of example, to evaluate a degree of modulation in a predetermined frequency range using speech modulation between 1 hertz and 5 or 10 hertz for speech recognition.
In one conceivable embodiment of the method according to the invention, the correlation between the first frequency band and the reference band is ascertained on the basis of signal amplitude or signal energy of a signal in the first frequency band and reference band.
The signal amplitude or the signal energy in the first frequency band and the reference band can be taken as a basis for ascertaining the correlation in a particularly simple manner.
In one possible embodiment of the method according to the invention, the correlation between the first frequency band and the reference band is ascertained on the basis of a degree of modulation of a signal in the first frequency band and reference band in a predetermined frequency range. By way of example, it is thus conceivable, in the method according to the invention, to use a degree of modulation having, in a predetermined frequency range, a speech modulation between 1 hertz and 5 or 10 hertz for determining the correlation.
High modulation is characteristic of speech. Therefore, the correlation of the instantaneous degree of modulation or the degree of modulation determined over a concurrent window affords reliable recognition of whether there is a speech component in the first frequency band too.
In one conceivable embodiment of the method according to the invention, the correlation is ascertained over a window length on the basis of the first audio signal.
The correlation must always be ascertained over a certain number of values that indicate a certain period in the waveform of the signals. This period or the number of values is also called window length. In this case, it may be advantageous to alter this window length on the basis of the environment and hence the first audio signal. By way of example, the fast changes in the input signal thus require the window length to be reduced in order to be able to react more quickly to changes of environment with the noise suppression.
In one possible embodiment of the method according to the invention, a value for a speech component of a signal is ascertained by comparing the correlation with a predetermined threshold value.
The reference band is selected in the method according to the invention such that it preferably has a speech component. If the first frequency band has a correlation with a sufficiently high value, that is to say that the reference band and the first frequency band have features sufficiently common, it can advantageously be assumed that the first frequency band also has speech components.
In one conceivable embodiment of the method according to that invention, a parameter of the noise suppression is set that is a parameter for influencing a spurious signal estimate, a parameter for setting the level of the spurious signal suppression or a parameter for limiting a spurious signal suppression.
Advantageously, the noise suppression in a frequency band can thus be set on the basis of an established speech component such that noise is preferably suppressed and speech components remain as uninfluenced as possible.
The embodiments of the hearing apparatus according to the invention on which the method according to the invention can be carried out share the advantages of this method.
Other features which are considered as characteristic for the invention are set forth in the appended claims.
Although the invention is illustrated and described herein as embodied in a method and an apparatus for noise suppression based on inter-subband correlation, it is nevertheless not intended to be limited to the details shown, since various modifications and structural changes may be made therein without departing from the spirit of the invention and within the scope and range of equivalents of the claims.
The construction and method of operation of the invention, however, together with additional objects and advantages thereof will be best understood from the following description of specific embodiments when read in connection with the accompanying drawings.
Referring now to the figures of the drawings in detail and first, particularly to
Furthermore, the hearing aid 110 according to the invention has a noise suppression device 20 that, as shown in
One possible embodiment of a noise suppression device 20 is shown in more detail in function blocks in
In the embodiment shown in
A reference band selection device 23 selects a reference band from the plurality of frequency bands. To this end, the reference band selection device 23 indicated in
A sub-band correlation determination device 24 is supplied with the signal in the reference band and also at least in a different frequency band, the connection for signaling purposes not being shown in
As a result, the sub-band correlation determination device 24 preferably delivers, by means of the speech identification device 26, a value for a probability of the signal in the first frequency band having a speech component. Alternatively, or additionally, the sub-band correlation determination device 24 can also use a decision device 25 to deliver a binary signal for whether the first frequency band has a speech signal. It is thus possible to simplify the subsequent handling when the first frequency band has no speech components.
A parameter setting device 27 takes the binary signal and/or the probability value as a basis for ascertaining suitable parameters for an adaptive noise filter 28. In the simplest case, the gain of the adaptive noise filter 28 can be set to zero if there is no speech component in the first frequency band. It is also conceivable for the gain to be proportional to or otherwise dependent on the probability value. It is also possible for other parameters of the adaptive noise filter 28 to be set in another way on the basis of the binary signal and/or the probability signal. It would thus be possible to adjust the step size for adaptive adjustment of the filter, e.g. a Wiener filter.
The adaptive noise filter 28 takes the parameter setting as a basis for preferably reducing a component of noise in comparison with speech components in the first frequency band.
A step S10 involves an audio signal being split into a plurality of essentially disjunct frequency bands.
In this case, the audio signal preferably originates from one or more microphones, but the source may also be an electrical signal input or a wireless transmission device, for example. The splitting into subbands can be effected using one or more filter devices, for example a filter bank, a plurality of discrete filters or through transformation into a frequency space. In this case, essentially disjunct is intended to be understood to mean that the individual frequency bands overlap only to a small degree or not at all, for example by no more than one quarter, one tenth or one twentieth of their bandwidth.
A step S20 of the method according to the invention involves a first frequency band for the plurality of frequency bands being selected as a reference band, which has an establishable first component of a speech signal.
In this case, it is conceivable for the hearing aid to have a device for recognizing a speech component. This device may be the reference band selection device shown in
A step S30 of the method according to the invention involves the hearing aid ascertaining a correlation between the first frequency band and the reference band. By way of example, this can be affected using the subband correlation determination device 24. Methods for determining a correlation are specified in the description relating to
Usually, the first frequency band is a different frequency band than the reference band. In the case of parallel or sequential handling of all the frequency bands by means of steps S30 to S50, however, it is also possible for the first frequency band to be identical to the reference band. The correlation value is then at a maximum owing to the identity.
A step S40 of the method according to the invention involves the decision device 25 and/or the speech identification device 26 taking the ascertained correlation as a basis for determining a value that indicates a second component of a speech signal in the second frequency band. This may be a binary value, a “binary mask” or even a fuzzy value, for the purposes of fuzzy logic, which indicates a probability. In this case, it is conceivable for a strong correlation between the first frequency band and the reference band to be able to be taken as a basis for inferring that the first frequency band also has a signal component having speech. However, it is also conceivable for the first frequency band likewise to be rated using a device for recognizing a speech component. It is also possible for the speech identification device 26 to be applied only to the first frequency band if sufficiently high correlation with the first frequency band has been ascertained and the decision device 25 assumes a corresponding value as “binary mask”.
A step S50 of the method according to the invention involves a noise suppression being set in the first frequency band on the basis of the ascertained value. In the simplest case, the gain of the noise filter 28 can be set to zero if there is no speech component in the first frequency band. It is also conceivable for the gain to be proportional to or otherwise dependent on the probability value. It is also possible for other parameters of an adaptive noise filter to be otherwise set on the basis of the binary signal and/or the probability signal. It would thus be possible to adjust the step size for adaptive adjustment of the filter, e.g. a Wiener filter, or the maximum possible cut.
The adaptive noise filter takes the parameter setting as a basis for preferably reducing a component of noise in comparison with speech components in the first frequency band.
Preferably, steps S30 to S50 are carried out for all the frequency bands from the plurality of frequency bands either in parallel or sequentially one after the other.
In a preferred embodiment, according to the invention, of the method, steps S10 to S50 are repeated cyclically at predetermined or variable intervals of time, step S20 being able to be carried out only once or likewise being repeated with a variable reference band. Within these repetitions, steps S30 to S50 are then preferably carried out for a plurality of or all the frequency bands in parallel or sequentially.
Although the invention has been illustrated and described in more detail by the preferred exemplary embodiment, the invention is not restricted by the disclosed examples and other variations can be derived therefrom by a person skilled in the art without departing from the scope of protection of the invention.
| Number | Date | Country | Kind |
|---|---|---|---|
| 102015201073.2 | Jan 2015 | DE | national |
