Patent Grant
8515107
This application claims priority of German application No. 10 2007 033 877.7 DE filed Jul. 20, 2007, which is incorporated by reference herein in its entirety.
The invention relates to a method for processing input signals in a hearing aid and to an apparatus of a hearing aid for processing input signals.
The enormous advance in microelectronics nowadays allows extensive analog and digital signal processing, even in the most confined spaces. Over the last few years, the availability of analog and digital signal processors with minimal spatial dimensions has also smoothed the way for their use in hearing aids, obviously a field of application in which the system size is subjected to a significant restriction.
In the case of hearing aids, a simple amplification of an input signal from a microphone often leads to an unsatisfactory result, since interference signals can be amplified at the same time and the benefits for the user can be restricted to special acoustic situations. Digital signal processors which digitally process the signal of one or several microphones in order thus to suppress interference noises in a targeted fashion for instance have therefore already been integrated into hearing aids for a few years.
Modern signal processing methods may include inter alia a so-called blind source separation (Blind Source Separation, or BSS for short), with several acoustic sources being broken down into individual signals. To this end, two or several microphones can record the acoustic environment and provide corresponding input signals for further processing. A classification of the input signal and/or of the input signals is also known, with a placement of the actual acoustic situation taking place on the basis of classification variables, like for instance of the input signal level or the number of detected acoustic sources. The source separation can then be adjusted according to the determined signal situation in order to provide the user of the hearing aid with an optimum output signal.
Methods for blind source separation, also known in brief as source separation, can include here a filtering of two input signals for generating two output signals and the determination of a so-called cross correlation, also known in brief as correlation, of the two output signals. The filters are adjusted until the correlation reaches a minimum, which then corresponds to a maximum separation of the output signals. The filtering generally takes place with the aid of a coefficient set, which includes at least one coefficient. A coefficient set, for instance in the form of a scalar, vector or matrix, is then used to filter the corresponding input signal.
In certain signal situations, a source separation can however achieve a local minimum of the correlation, although an absolute minimum exists. The source separation then erroneously results in a maximum separation already taking place and omits a significant change in the coefficient set. In such a case, this is also referred to as freezing the source separation. A frozen source separation is not able to change the coefficient set further although the correlation has still not reached an absolute minimum and the source separation thus does not take place completely.
In order to identify such a disadvantageous freezing, an additional source separation can take place for instance, for example a second source separation or a so-called shadow source separation, which is continuously reset in order to avoid freezing this additional source separation. If such an additional source separation determines a smaller correlation than the main source separation, for instance a first source separation, which is not regularly continuously reset, this can be an indication of the main source separation being frozen in a local minimum.
In such an instance, the coefficient set of the main source separation can be replaced at least partially by the coefficient set of the second source separation. The disadvantage here is that since its last reset and/or initialization until reaching a smaller correlation than the main source separation, the additional source separation has only minimally adapted the corresponding coefficient set. Since this inadequately adapted coefficient set then determines the coefficient set of the main source separation, the main source separation can deliver unsatisfactory results here.
DE 10029388 A1 discloses a signal separation apparatus, in which a first and a second signal of the signal sources originate from two systems, which are transmitted to a receiving device provided with two receiving facilities. The first and the second signal mutually superimpose one another as a result of crosstalk during their transmission through a transmission channel section. The two signals are received by the two input facilities of the receiving device. The signal separation apparatus has a signal separation section and an evaluation function calculation section. The signal separation section has six filter facilities with variable branching coefficients. The evaluation function calculation section has a first and a second autocorrelation calculation facility and a minimum value determination facility for determining a minimum value.
EP 1655998 A2 discloses a method for generating stereo signals for separated sources and a corresponding acoustics system. Here a blind source separation is carried out by at least two microphone signals for obtaining transmission functions of filters of a first filter facility. Transmission functions of filters of a second filter facility are also determined with the aid of the transmission function of the filter of the first filter facility. The two microphone signals are filtered in each instance using at least two filters of the second filter facility.
WO 00/25489 A1 discloses a method and an arrangement, which enables the separation of superimposed acoustic signals, which are statistically independent of each other, with a reduced computing outlay. Here parameters of a technical system are determined such that the statistical independence of the output signals is maximized. The parameters are elements of a separation matrix, with which the quantity of superimposed input signals is multiplied or also folded, as a result of which the output signals are formed. With the optimization of the parameters of the separation matrix, the following steps are implemented:
Repeating a time-delayed decorrelation calculation for determining the intrinsic values of the separation matrix;
Determining the intrinsic values of the separation matrix for which the correlations assume a minimum value;
Implementing a cumulant minimization, with the intrinsic values determined in the previous step being used as start values for the cumulant minimization. A linear filtering can result in a Gauss distribution with distortion of the data, in which the cumulants of a higher order move toward 0. This can thus result in marginal solutions, in which the cost function reaches a local minimum, with the desired actual separation (global minimum) not taking place. To avoid this unwanted situation, the structure of the separation transfer function is subjected to some restrictions.
It is thus the object of the present invention to provide an improved method for processing input signals in a hearing aid, as well as an improved apparatus for processing input signals in a hearing aid.
This object is achieved by the method and by the apparatus as claimed in the independent claims.
Additional advantageous embodiments of the invention are specified in the dependent claims.
According to one first aspect of the present invention, provision is made for a method for processing two input signals in a hearing aid, with the input signals being dependent on an acoustic signal, which includes the following steps: a first source separation, with two first output signals and a first correlation of the two first output signals being continuously determined from the two input signals with a first coefficient set and with the first coefficient set being continuously determined as a function of the first correlation; a second source separation, with two second output signals and a second correlation of the two second output signals being continuously determined from the two input signals with a second coefficient set, and with the second coefficient set being continuously determined as a function of the second correlation; a comparison of the first correlation with the second correlation and a change in the first coefficient set as a function of the second coefficient set, if the second correlation is smaller than the first correlation, with the method also including changing the second coefficient set as a function of the first coefficient set when resetting the second source separation.
According to a second aspect of the present invention, provision is made for an apparatus for processing two input signals which are dependent on an acoustic signal in a hearing aid, with the apparatus having a processing unit, which continuously determines two first output signals and a first correlation of the two first output signals from the two input signals with a first coefficient set and continuously determines the first coefficient set as a function of the first correlation and continuously determines two second output signals and a second correlation of the two second output signals from the two input signals with a second coefficient set and continuously determines the second coefficient set as a function of the second correlation, with the processing unit changing the first coefficient set as a function of the second coefficient set, if the second correlation is smaller than the first correlation and with the processing unit resetting the second coefficient set as a function of the first coefficient set.
The method according to the invention and the apparatus according to the invention are advantageous in that the second source separation can start with a reset with a coefficient set, which is dependent on the first coefficient set of the first source separation. Here the second source separation can revert back to a coefficient set after a reset, which can advantageously already include an adaptation. This adaptation can be contained in the first coefficient set by means of the first source separation. The second source separation can thus omit part of an adaptation, which was already accounted for in the corresponding first coefficient set and can advantageously achieve a corresponding adaptation more rapidly.
If in the case of a freezing the first coefficient set of the first source separation is changed as a function of the second coefficient set of the second source separation, the first coefficient set can have an already significantly increased adaptation and can thus result significantly more rapidly in a satisfactory result for the user of the hearing aid.
According to one embodiment of the present invention, the method includes storing the first coefficient set. This storage can take place as a function of a temporal change in the first correlation. The first coefficient set can also be stored, if the first correlation within a first period of time is still only marginal, with it being possible for the first period of time to lie within a range of 1 second to 10 seconds. A marginal variation in the first correlation can be provided if the correlation does not vary by a value of more than 30%, more than 10% or more than 5%.
Accordingly, an advantageous first coefficient set can be stored and is then available at the time of resetting the second source separation in order to change the second coefficient set accordingly as a function of the stored first coefficient set. Advantageous first coefficient sets can thus be stored again if the first correlation changes temporally minimally. A minimal temporal change in the first correlation can be an indication that the first source separation provides a satisfactory result and performs an optimum source separation which corresponds to the respective acoustic signal situation.
According to a further embodiment of the present invention, the storage of the first coefficient set can take place in a storage unit of a group of storage units, with the selection of the storage unit taking place as a function of a signal situation. Various first coefficient sets can thus be stored for several signal situations.
According to a further embodiment of the present invention, at least two first coefficient sets for a signal situation are stored in the storage units of the storage unit group. At least two predetermined coefficient sets can thus be recalled from the storage units as a function of the signal situation and the first coefficient set is changed as a function of the predetermined coefficient sets. That predetermined coefficient set, for which a change in the first coefficient set produces a minimum first correlation, is also determined. Several coefficient sets are thus advantageously available for a signal situation, from which coefficient sets that coefficient can be selected, for instance by a comparison of the respectively resulting correlation, which produces a minimum correlation and thus an optimum separation efficiency.
According to a further embodiment of the present invention, the second source separation is reset after a second period of time has elapsed. The second source separation can herewith be periodically reset and a potentially occurring, disadvantageous freezing of the second source separation can be prevented. Examples of a second period of time are periods of times in a range of 10 milliseconds to 10 seconds.
According to a further embodiment of the present invention, the second source separation is reset once a second period of time has elapsed. At least two first coefficient sets for a signal situation are also stored in the storage units of the storage unit group. At least two predetermined coefficient sets can thus be recalled from the storage units as a function of the signal situation and the second coefficient set is changed during the reset process as a function of the recalled coefficient sets. That recalled coefficient set, for which a change in the second coefficient set produces a minimum second correlation, is also determined. Several coefficient sets are thus advantageously available for a signal situation, from which that coefficient set can be selected, for instance by a comparison of the respectively resulting correlation, which produces a minimum correlation and thus optimum separation efficiency.
According to a further embodiment of the present invention, the second source separation is reset after a second period of time has elapsed and the second coefficient set is changed during the reset process as a function of a predetermined coefficient set. Here the predetermined coefficient set is recalled from a storage unit of the group of storage units as a function of a signal situation. The second source separation can thus advantageously also be reset according to the signal situation. In this way the second source separation can thus revert back to an output coefficient set which optimally corresponds to the actual signal situation after the reset process.
According to a further embodiment of the present invention, at least one of the following classification variables is determined: a level of an input signal, a level of an output signal, a distribution of the level, a performance spectrum or a spatial position of a source of one of the input signals. The signal situation can then be determined according to at least one of these classification variables.
According to a further embodiment of the present invention, the method includes a third source separation, with two third output signals and a third correlation of the two third output signals being continuously determined from the two input signals with a third coefficient set, with the third coefficient set being continuously determined as a function of the third correlation, with a comparison of the first correlation with the third correlation taking place and with the first coefficient set being changed if the third correlation is smaller than the first correlation.
A method according to this embodiment of the present invention can advantageously also capture a possible freezing of the second source separation. As the third source separation can be reset with a universal output coefficient set, this provides a third correlation which is independent of the signal situation and the already adapted coefficient sets. The third source separation is thus not prone to freezing, which can in some circumstances arise in special combinations with the first and second coefficient set and/or a corresponding situation. In such an instance, the first source separation can then be advantageously reset and the first coefficient set can be set as a function of the third coefficient set.
Preferred embodiments of the present invention are described in more detail below with reference to the appended drawings, in which;
A correlation module 172 determines a first correlation of the two first output signals 111, 112 from the first output signal 101 and from the additional first output signal 112. The result of the correlation module 172 is fed back to the filter module 171, so that the filter module 171 correspondingly modifies internal filters, for instance in the form of a coefficient set, in order to achieve a minimization of the first correlation of the two first output signals 111, 112. If the source separation 100 has achieved an absolute minimum of the correlation, the output signals 101, 112 have a minimum correlation and are thus maximally separated.
If the first input signal 101 has passed through the first filter 141 and the second input signal 102 has passed through the second filter 142, these signals are added to the first output signal 111. Similarly here, the additional first output signal 102 is provided by adding the first input signal 101, which has passed through the third filter 143 and by adding the second input signal 102, which has passed through the fourth filter 144. The correlation module 172 determines the correlation of the two first output signals 111, 112, and correspondingly controls the filters 141, 142, 143, 144, thus striving for a corresponding minimum of the correlation.
The first source separation 210 includes a filter module 211 and a correlation module 212. The filter module 211 generates a first output signal 111 as well as an additional first output signal 112 from the first input signal 101 and the second input signal 102 with the aid of a first coefficient set W1. The correlation module 212 continuously determines a first correlation κ1 from the two first output signals 111, 112, which is used to modify the first coefficient set W1 such that the first correlation κ1 is minimized and the two first output signals 111, 112 are thus maximally separated.
A second source separation 220 includes a filter module 221 and a correlation module 222. The filter module 221 generates a second output signal 121 as well as an additional second output signal 122 from the first input signal 101 and the second input signal 102 with the aid of a second coefficient set W2. The correlation module 222 continuously determines a second correlation κ2 from the two second output signals 121, 122, which is used to modify the second coefficient set W2 such that the second correlation κ2 is minimized and the two second output signals 121, 122 are thus maximally separated.
According to this embodiment of the present invention, a comparison module 241 compares the first correlation κ1 with the second correlation κ2. If the second correlation κ2 is smaller than the first correlation κ1, the comparison module 241 changes the second coefficient set W2 as a function of the first coefficient set W1.
To this end, the comparison module 241 can revert back to stored coefficient sets W1A, W1B, etc., which are stored in a storage unit group 243. The coefficient sets W1A, W1B, etc., which are stored in the storage unit group 243, can be in correspondingly different signal situations dependent on the first coefficient set W1. Provision can thus be made to store an adapted coefficient set W1 in a first signal situation “A” as a coefficient set W1A in the storage unit group 243. Provision can also be made to store the first coefficient set W1 in a second signal situation “B” as a coefficient set W1B in the storage unit group 243 etc.
Provision can also be made for the comparison module 241 to also revert back to several stored coefficient sets W1A1, W1A2, . . . , W1B1, W1B2, . . . for a signal situation, which are stored in a storage unit group 243. The coefficient sets W1A1, W1A2, . . . , W1B1, W1B2, . . . can be stored again in the storage unit group 243, and are stored there for instance in the event of a change in the corresponding correlation which is still only marginal. Provision can thus be made for several adapted coefficient set W1 in a first signal situation “A” to be stored as coefficient sets W1A1, W1A2, . . . in the storage unit group 243. Several coefficient sets W1 are thus available for a signal situation, from which that coefficient set W1, for which a minimum first correlation and thus an optimum separation efficiency results, can be selected.
Acoustic signal situations “A”, “B”, etc., known to the hearing aid, which are to imitate situations in daily life, can be assigned to an actual signal situation on the basis of corresponding classification variables for instance. To determine the actual signal situation of the most similarly known signal situation, a determined classification variable does not have to be identical to a classification variable of the known signal situations, but bandwidths and tolerances can instead be provided for the respective classification variables for instance. Examples of known acoustic signal situations “A”, “B”, etc., are indicated in the following table:
Here the signal components can be the output signals, the input signals or can also be generated from an additional decomposition and/or separation of the input signals and/or output signals. An additional apparatus and/or additional method can be provided in the hearing aid in order to determine a signal component, a signal/noise ratio, a performance spectrum, a level, a number of signal components, an additional classification variable and/or the signal situation.
The comparison module 241 can also assign a corresponding signal situation and use one of the coefficient sets W1A, W1B etc. stored in the storage unit group 243 etc., in order to change the first coefficient set W1. To this end, the comparison module 241 can determine a corresponding signal situation, or receive the correspondingly determined signal situation from an additional module, a method or an additional apparatus. Provision can also be made for the differentiation to be omitted in different signal situations and for the storage unit group 243 only to store a coefficient set as a function of the first coefficient set W1. In this instance, the storage unit group 243 can be replaced by an individual storage unit.
According to this embodiment of the present invention, a timer module 244 resets the second source separation 220 once a certain period of time has elapsed, for instance once a second period of time has elapsed, by the timer module 244 setting the second coefficient set W2 as a function of a first coefficient set W1A, W1B, etc. To this end, the timer module 244, can select a first coefficient set W1 from the storage unit group 243 according to the current signal situation, and can then reset the second source separation 220 by means of correspondingly setting the second coefficient set W2. Here the timer module 244 can take over a complete coefficient set W1A, W1B, etc., in the second coefficient set W2, or only a part thereof. For example, the storage depth of the storage unit group 243 can allow the storage of 1024 coefficients for each coefficient set W1X (X=A, B, etc.), with only a part of the coefficients, for instance 512 coefficients, being taken over as coefficient set W2 during the reset process.
Provision can also be made for the second coefficient set W2 also to be changed for a signal situation also as a function of several stored coefficient sets W1A1, W1A2, . . . Several coefficient sets W1 are thus available for a signal situation, from which that coefficient set W1, for which a minimum second correlation and thus an optimum separation efficient result, can be selected.
According to this embodiment of the present invention, the second source separation 220 is as a result reset with an already adapted coefficient set and/or with a coefficient set adjusted to a current signal situation. Advantageously the second source separation 220 can thus already begin with an at least partially adapted coefficient set. As a result, the second source separation 220 can perform a source separation more quickly and the corresponding second correlation κ2 can drop more quickly below the first correlation κ1 in the event that the first source separation 210 freezes and the first source separation can be advantageously reset more rapidly, it can thus respond more rapidly to a disadvantageous freezing of the first source separation 210.
By resetting the second coefficient set W2 as a function of one of the first coefficient sets W1A, W1B, etc. in the instance of the first source separation 210 of the first coefficient W1 freezing as a function of the second coefficient set W2, being at least partially reset, with it being possible at this time for the second coefficient set W2 to advantageously comprise an already progressed adaptation. After determining and after correspondingly eliminating the freezing of the first source separation 210, the first source separation 210 can thus continue with an at least partially adapted first coefficient set W1, and can thus provide a satisfactory output signal and result for the user of the hearing aid more rapidly.
According to this second embodiment of the present invention, a third source separation 230 is also provided. Here the third source separation 230 includes a third filter module 231 as well as a third correlation module 232. The filter module 231 generates a third output signal 131 as well as an additional third output signal 132 from the first input signal 101 and the second input signal 102 with the aid of a third coefficient set W3. The correlation module 232 continuously determines a third correlation κ3 from the two third output signals 131, 132, which is used to modify the third coefficient set W3 such that the third correlation κ3 is minimized and the two third output signals 131, 132 are maximally separated. An additional timer module 245 is also provided which resets the third source separation 230 once a certain period of time, for instance a third period of time has elapsed, by the third source separation 230 setting the third coefficient set W3 to an output coefficient set.
According to this embodiment of the present invention, the first correlation κ1, the second correlation κ2 and the third correlation κ3 are fed to an additional comparison module 242, which compares the first correlation κ1 with the second correlation κ2. The additional comparison module 242 can also compare the first correlation κ1 with the third correlation κ3 and/or the second correlation κ2 with the third correlation κ3. If the second correlation κ2 is smaller than the first correlation κ1, the additional comparison module 242 changes the second coefficient set W2 as a function of the first coefficient set W1, as was already described in conjunction with
If the third correlation κ3 is smaller than the first correlation κ1 and/or the third correlation κ3 is smaller than the second correlation κ2, this can be an indication that the second source separation 220 is frozen. It is also possible that the first source separation 210 is frozen. As the third source separation 230 can be reset with a universal output coefficient set, this provides a third correlation κ3, which is independent of the signal situation and the already adapted coefficient sets. In such an instance, the first source separation can then be advantageously reset and the first coefficient set is reset as a function of the third coefficient set. The same can also occur for the second source separation 220.
According to embodiments of the present invention, modules, like for instance the filter modules 211, 221, 231, the correlation modules 212, 222, 232, the comparison modules 241, 242, and/or the timer modules 244, 245 can be embodied both as discretely embodied circuits as well as processes, for instance as a thread or task, running in a microprocessor, in a signal processor, or in an integrated process module.
According to this embodiment of the present invention, the additional processing unit 430 has a process unit 440, which includes processes or modules, like for instance a first module 441, a second module 442, a third module 443 etc. The modules 441, 442, 443 can include for instance a source separation, a comparison module, a timer module, a filter module and/or a correlation module, like was illustrated in conjunction with the embodiments of the present invention described in
If in the bifurcation 612 it is determined that the correlation κi is not minimal, the coefficient set Wi is changed in a step 613. The coefficient set Wi can correspond here to the first coefficient set W1, the second coefficient set W2 or the third coefficient set W3, like was described in conjunction with
It is determined in a bifurcation 634 whether the temporal change in the first coefficient set W1 within the first period of time is below a threshold value. This threshold value can correspond to a characteristic threshold value for instance, which delimits an adapted coefficient set W1 from a non-adapted coefficient set W1. If the temporal change in the first coefficient set W1 lies below the threshold value, it is thus possible to determine for instance that the current coefficient set W1 corresponds to a well adapted coefficient set for a special signal situation. Hereafter the corresponding first coefficient set W1 can be stored in a following step 635 as a coefficient set for the corresponding signal situation. If the temporal change in the coefficient W1 does not lie below the threshold value, the process is continued with a renewed querying of the coefficient W1 (631).
The module according to this embodiment can also include buffering the coefficient set W1 in step 631, in order to compare the buffered coefficient set W1 with the coefficient set W1 determined in step 633 and/or to determine a temporal change in the coefficient set. According to the seventh embodiment of the present invention, the module can also correspond to a setting of the coefficient sets in the group of storage units for corresponding signal situations.
However, if the second correlation κ2 lies above the first correlation κ1, a freezing of the first source separation is not determined. In this instance, the process is continued by querying the first correlation (641).
The modules, as described in conjunction with
| Number | Date | Country | Kind |
|---|---|---|---|
| 10 2007 033 877 | Jul 2007 | DE | national |
| Number | Name | Date | Kind |
|---|---|---|---|
| 6704369 | Kawasaki et al. | Mar 2004 | B1 |
| 7043030 | Furuta | May 2006 | B1 |
| 7257231 | Avendano et al. | Aug 2007 | B1 |
| 20060120535 | Puder | Jun 2006 | A1 |
| Number | Date | Country |
|---|---|---|
| 10029388 | Mar 2001 | DE |
| 1655998 | May 2006 | EP |
| WO 0025489 | May 2000 | WO |
| Number | Date | Country | |
|---|---|---|---|
| 20090022344 A1 | Jan 2009 | US |
