EP2792165A1

EP2792165A1 - Adaptation of a classification of an audio signal in a hearing aid

Info

Publication number: EP2792165A1
Application number: EP12701891.9A
Authority: EP
Inventors: Roland Barthel; Marko Lugger
Original assignee: Siemens Medical Instruments Pte Ltd
Current assignee: Sivantos Pte Ltd
Priority date: 2012-01-27
Filing date: 2012-01-27
Publication date: 2014-10-22
Anticipated expiration: 2032-01-27
Also published as: DK2792165T3; US20140369510A1; AU2012367084B2; WO2013110348A1; EP2792165B1; AU2012367084A1; US9294848B2

Abstract

The invention relates to the adaptation of the classification of an audio signal (8) as a function of a comparison of two difference sums of audio features over time periods of different length. Thus, an adequately exact yet quickly reacting adaptation of the classification in changing hearing situations is ensured. The method according to the invention is advantageously used in a hearing aid (13). The audio signal (8) is processed in different ways on the basis of the classification.

Description

description

Adaptation of a classification of an audio signal in a hearing aid

The present invention relates to a process for ANPAS ^¬ solution of a classification of audio signals. Moreover, the present invention relates to a corresponding signal processor and a hearing aid.

Hearing devices serve primarily to make audio signals of sound waves for a particular desired purpose easier to understand. A sub-area of the use of hearing devices as a hearing aid is the care of the hard of hearing. The amplification process of a hearing device runs off by means of the built-in electronics. One or more microphones on the hearing device receive an audio signal, which is processed by means of an audio processor and output by a listener.

Depending on the whereabouts of the user of a hearing device he ^¬ different hearing situations. In many listening situations, such as a car ride, desirable and unwanted sounds occur. In the example of the drive-by noise of Au ^¬ tos is desired, the voice of a passenger, undesirable. Preferably, only desired sounds are to be filtered out and further processed by a hearing device. Frequently occurring hearing situations can be classified. This classification is performed by a Signalpro ^¬ cessor which assigns a certain classification of one or more kinds of audio features of an audio signal using an algorithm, this audio signal. An Au ^¬ diomerkmal example, level or amplitude of an Au ^¬ diosignals be. Furthermore, an audio processor with the information of a classification corresponding to this can further process the audio signal. An audio processor has various processing programs that are selected depending on the Klassifika ^¬ tion. The process for setting a classification is fundamentally shaped by two requirements. Best fit single note first, a classi ^¬ fication of the current hearing situation and second, the speed of this time

Attitude. Accuracy and a quick change of classification are in competition.

It is an object of the present invention to enable a responsive change of classification depending on a ge ^¬ changed listening situation under warranty tung a reliably stable classification.

This object is achieved by a method for adapting a classification of an audio signal according to claim 1, by a method for classifying an audio signal according to claim 9 and by a hearing device according to claim 12.

By comparing difference sums of audio features that are summed over different lengths of time, short-term changes in the provided audio signal can be reliably identified with respect to a longer observation period, providing a reliable basis for

Performing a classification change is formed. The classification change is based on the temporal sequence of differences of successive values of an audio feature of the audio signal, so that the change is taken into account in the form of a plurality of intermediate values over a certain period of time, whereby a change of the hearing situation is certainly reflected in the differences of the feature values.

On one hand, a change in the audio signal is fast identifi ed ^¬ by the observation of a first, shorter period of time and on the other hand, a sufficient stability of the classical sification is guaranteed by reference to a two ^¬ th, longer period.

An audio feature is a quantity derived from an audio signal. Typically, the audio feature refers to timed Liehe aspects, ie phase or frequency, or on the Ampli ^¬ tude an audio signal. The audio feature thus also changes over time according to the audio signal. In width ^¬ ren the audio feature, an average, a standard deviation ^¬, a modulation or variation of a level of the audio signal can be.

According to a development, the comparison is carried out by means of a quotient of the first sum and the second sum. A quotient is easily determined by a simple mathematical operation and provides a meaningful measure of the ratio of the first sum and the second sum.

According to a development, a temporal sequence of values of different types of audio features is generated and the difference of individual differences is formed by the successive values of similar audio features. The audio features may be averages, standard deviations, modulations or variances of a level of an audio signal. To use Various audio features instead of loading ^¬ restriction to a specific audio feature ver ^¬ improves the accuracy of the classification. The Einzeldiffe ^¬ narrow be weighted in the formation of the difference on the type of each audio feature, which increases the flexibility of determining a classification change in the claimed process.

According to a development, the values of the various audio features are combined into a feature vector and the difference is obtained in the form of a distance between successive feature vectors. Summarizing a vector makes it easier to process the audio features.

In a further development of the classification change is from ^¬ pending conducted by a currently selected classification. By a further dependence of the classification ^¬ change also from a currently selected classification can Classification depending on the stability or the responsiveness of a classification change controlled ^¬ the. For example, the classification change from a listening situation for speech can take place only if the comparison indicates the total difference in the sequence of audio features particularly evident on changing the listening situation to target an increased stability for the class of language he ^¬. Advantageously, the first period has a duration of 2 to 6

Seconds and the second period lasts 10 to 20 seconds.

It is further contemplated to provide a method for classifying an audio signal, which preparation comprises the steps of a ^¬ initially mentioned method, and additional steps before ^¬ a classification change by selecting a pre ^¬ shock for a customized classification in dependence on a value of the audio feature and implementing the classification change according to the proposal for an adapted classification depending on the comparison.

It is made a concrete proposal for a classification ^¬ change. The additional presence of such a proposal shortens the time to change into a classi ^¬ fication, as can be ge ^¬ changes based on the proposal in a classification without having to complete the calculation for the classification change. The present invention will now be explained in more detail with reference to Ausführungsbei ^¬ play in the accompanying drawings, in which:

1 shows the flow of a method for adapting the classical sification of an audio signal according to a Ausges ^¬ taltung the invention. FIG. 2 shows the time profile of an audio signal and, in relation thereto, the time periods relevant for the method according to FIG. 1; FIG.

Figure 3 shows the sequence of the method according to Figure 1 in together ^¬ menhang with a classification change.

4 shows a hearing aid with a signal processor for carrying out the method according to FIG. 3; and

5 shows a magnification view of the signal processor from the hearing aid according to FIG. 4.

FIG. 1 schematically shows the sequence of a method according to an embodiment of the present invention. This method can e.g. run in a signal processor of a hearing aid.

In a first step 1, an audio signal is provided. This audio signal is typically a microphone ^¬ signal of the hearing aid. The microphone signal may be supplied by one or more microphones of the hearing aid. Furthermore, between the microphone or the microphones and the signal processor, a signal processing can be interposed, which, for example, smoothes the microphone signal.

In a second step 2 we generate a temporal sequence of values _{k of} an audio feature. The values of the sequence are in this case numbered by an index k in a chronological order. Not just a single audio feature is considered advantageous, but several ver ^¬ different type. In this case, u _k represents a shopping times ^¬ vector containing the values of these audio feature to the index k corresponding time t _k sums. The time ^{interval of} two successive times t _k _ _l and t _k may be, for example, between 10 ms and 200 ms. The audio feature ^¬ represents characteristics of the Au- diosignals at a certain time. The audio feature is typically determined from the timing of the audio signal in a temporal environment around the particular time. The skilled person is known per se, for example, a mean, a standard deviation, a modulation or variation of a level of Audiosig ^¬ Nals various audio features.

In a third step 3, a difference ^u _k ^{~ u} _k - _{\ is} formed from successive values u _k _ _l and u _{k of} the audio features. For the different values k = 1,2,3, etc. of the index is thus obtained in this way a series of conferences Dif ^¬. For the subsequent method steps is essentially the absolute value of this different, ie

d _k = \ ^u _k ^{~ u} _k -iI decisive. In case of a feature vector for egg ^¬ ne plurality of audio features _k d represents the distance of the successive vectors u _{k γ} and u _k. The distance can be chosen differently, eg as Euclidean distance or Mahalanobis distance. At this distance, the audio features may also be weighted differently, eg by multiplying the feature values by different scalar coefficients before determining the distance. In the following, _{dk will} be referred to as difference only, depending on

Embodiment also represent the absolute amount of the difference or the distance can stand.

In the next steps 4 and 5, the sequence of differences d _{k is} variously processed by summing them over periods of different lengths. In step 4, the differences over a first period T _x are added up to a first sum Σ ₁ . In step 5, however, the differences are summed over a longer period T ₂ to a second sum Σ ₂ . The shorter period T _x may be, for example, 2 to 5 seconds, and the longer period T _{2 may be} 10 to 20 seconds. The longer period T ₂ is in this embodiment, two to ten times longer than the shorter period T _x . For a shorter period T _x of, for example, 2 seconds and a time interval of successive values of the audio signals, for example, 10ms are thus for the period T _x 200 individual values of differences of the values u _k summed, the t times _k corresponding to that fall within the period T _x. The sum of the differences thus describes the set of all ^¬ Einzelände conclusions of the audio features on the respective sum period.

In a sixth step 6, the two sums Σ ₁ and Σ ₂ over prior periods T _x and T ₂ are compared with each other. By this comparison of the totality of the individual changes over two different periods of time striking short-term changes in relation to a County ^¬ gerfristigen trend are visible. The comparison is made in a simple manner by forming a quotient of Σ ₁ and Σ ₂ , whereby the relative length of the two periods must be taken into account in the evaluation of the quotient. For example, the value of the quotient

0 = ^ - ^ - with T _? > T,

Σ ₂ · Τ; be considered in the comparison. It is thus virtually the average rate of change _Σ ι / _Τ ι in the shorter time period T ^¬ _x with the average rate of change Σ ₂ / Γ _{2 are} compared in the longer period of time T _2. If the value of Q essential lent greater than 1, this indicates a striking raised stabili ^¬ hung the rate of change in the period T _x.

In a seventh step 7, a classification change is performed as a function of the comparison in the preceding step 6. Here is not the classification itself ge ^¬ selected, but only converted one proposed by another road classification. The classification proposal itself can be determined in a conventional way depending on the listening situation. By the present method a classification change is thus inhibited, if the comparison described above suggesting that the situation Hörsi ^¬ _x is not significantly altered in the past period T has. However, since the period Ί is chosen to be relatively short, this method allows a quick yet reliably determined classification change.

By taking into account various audio features, and possibly in addition by a weighting of these various audio features, the method can be tuned sensitive ^¬ the. The selection and the weighting can be improved, for example, by means of test series in various changing hearing situations with regard to an exact recognition of a change in the hearing situation.

In addition, the change of the classification can also be carried out depending on the currently selected listening situation. It is, for example, the example is desirable that Hörsi ^¬ situation "language alone," particularly stable against erroneous classification changes, whereas others Klassifikatio ^¬ nen like "Auto", "Music", "calm" or "noise" can be easily changed. This change may also depend on a suggestive new classification, so that, for example, a change to the listening situation "speech at rest" can be done very quickly. In addition, the current and / or proposed classification may also take into account the weighting of the audio features in forming the distances. The Summierungszeiträume _Τ γ and T ₂ Kings ^¬ nen depend to further improve on the current and / or proposed classification.

The "speech at rest" listening situation is when a person speaks in an otherwise quiet environment, and also includes classifications of the listening situations in a car ("car"), for music ("music"), in a quiet environment ( The hearing aid is also classified based on the audio signal, whereby the audio features mentioned above can also be taken into account depending on the respective hearing situation the listening situation adapted Hearing program for processing the audio signal determined. The audio signal processed with the respective hearing program is reproduced to the wearer of the hearing aid in an amplified form. The hearing program determines, for example, various frequency filters, possibly also frequency-dependent gain and the directivity of the microphones.

FIG. 2 schematically shows the time profile of an audio signal 8 and the relation of the individual time periods T _i and T _2i , over which the sequence of differences of the values of the audio features is summed. A period for k = -20 to k = +80 is shown. By way of example, every tenth time t _{k is} indicated on the horizontal time axis. The vertical one

Axis indicates the respective amplitude of the audio signal 8.

Below the time axis, a sequence of short Zeiträu men ^¬ T _{X i} and a further series of longer time periods T _2i is indicated. The short periods T _Xi each comprise ten individual intervals between the times t _k . Thus, the associated sums Σ _Η include the differences of ten consecutive value pairs u _k _ and u _k . The longer periods T _2i are each three times as long as the short periods T _i . The associated sums Σ _2ί thus comprise the differences of thirty consecutive value pairs u _{k γ} and u _k .

The numbering of the indices is chosen such that the ^intervals T _Xi and T _2i for the same index / ^'end at the same point in time t _k with £ = 10 · /. In this case, the period T _Xi always lies within the period T _2i , both ending at the same time. Alternatively, the T _{lt can} also connect directly to the period T _2i . In any case, the T _i and T _{2i should be} in close temporal relationship.

With each increment of the index / ^' the time intervals T _Xi and T _2i are the same amount, so that the relation of these intervals to each other is maintained. In this case, the time intervals T _Xi shift by the same duration Periods _Τ ίί so that the periods string together without time Un ^¬ interruption. Alternatively, it is also possible that the shift is larger or smaller than the time intervals T _j .

In the present case, the sums Σ _1; , and Σ _2ί are represented as an equation as follows:

From these sums, the following quotients Q _t can again be formed, on which the execution of the classification change depends: _ ^ i T _2J

As already described above, u _k can be both an individual

Its numerical value for a feature or a vector with a multi ^¬ number of individual values to various audio features. In the case of a single numerical value stands for Abso ^¬ lutbetrag. In the case of a vector, u _{k is} in the form of a tuple of numerical values where n is the index used to distinguish each numeric value. Different standards can be chosen depending on the application. ^One possible standard is the Euclidean norm, which is defined as follows.

The sum is made over all entries of the vector. Al ternatively ^¬ may be defined as Mahalanobis distance \ u _k -u _kl \. Figure 3 schematically shows the sequence of the method according to Fi gur ^¬ 1 in connection with a classification proposal. As in Figure 1, the sequence of steps represented as rectangles indicates a possible chronological order. Other sequences while maintaining the causal relationships are also possible.

In the illustrated embodiment, from the audio signal 8, a first value of an audio feature is created by ^¬ loading ^¬ woman on top of the audio signal at a first time in 8 to step 9. As before, instead of a single value, a multiplicity of values of different audio features can again be taken into account here. Based on this first value of the audio feature, a classification is selected in step 10. This choice is made according to well-known methods for classifying audio signals.

In later steps 11 and 12, both steps 9 and 10 described above are repeated. In step 11, at a second time a second value of Au ^¬ thus is diomerkmals generated which is based on a choice of a further classification. The now adjusted classification may differ from the previously selected classification. In such a case, the election classification corresponds to the second time the proposal for a classification change. However, this proposal will not be carried out initially.

In step 2, in the interval between the first time and the second time, the time sequence of the values of the audio feature is generated, as already described with reference to FIG. This sequence of values is shown in Fi gur ^¬ 1 the basis for the method steps 3 to 7. In

Step 7 depends on the actual implementation of the Klassifika ^¬ tion change from, the comparison of the two sums of differences, as described previously.

FIG. 4 shows a hearing device 13 with two microphones 14, an arrangement 15 of electronic components for signal processing. processing, a battery 16 and a receiver 17 for sound generation. The microphones 14 provide an audio signal 8. Through targeted signal processing can be generated with the two microphones 14 a directivity. The audio signal 8 is forwarded by electrical lines to the arrangement 15. The assembly 15 is powered by the battery 16 with electrical power. After the signal processing of the audio signal 8, the processed audio signal is passed to the handset 17 for output.

FIG. 5 shows a magnification view of the arrangement 13 of electronic components for signal processing from FIG. 4. The audio signal 8 from the microphones 14 arrives via an electrical contact 18 to an input ^interface 19 of a signal processor 20. A classification unit 21 in the signal processor 20 carries this with respect to Figure 3 described method for classifying the audio signal 8 by. The result of the classification is passed through a classification output 22 to an audio processor 23.

The audio processor 23 also receives the audio signal 8 directly from the microphones 14 via the contact 18. Based on the selected classification, the audio processor 23 processes the audio signal 8 by applying a classification program corresponding processing program adapted to the respective listening situation. The processed audio signal is sent by the audio processor 23 to the handset 17 of the

Hearing aid 13 on. An optionally intermediary amplifier for the processed audio signal is not shown for Vereinfa ^¬ chung in the drawing. Finally, the basic concept of at least one ^{embodiment of} the invention is summarized again. The invention relates to the adaptation of the classification of an audio signal as a function of a comparison of two differential signals. buzz audio features over varying lengths of time. In a sufficiently precise and yet Reaction ^¬ rapid adaptation of classification in varying listening environments is guaranteed. The inventive method is advantageously used in a hearing aid. Based on the classification, the audio signal is prepared in various ^ways .

Although the invention in detail by the preferred embodiment has been illustrated and described in detail, the invention is not limited by the disclosed examples is ^¬ limited and other variations can be derived therefrom by the skilled artisan without departing from the scope of the invention

Claims

claims

A method of adapting a classification of an audio signal (8), comprising the following steps:

Provision of the audio signal (8),

Generating a time sequence of values of an audio feature of the audio signal (8),

Formation of a chronological sequence of differences of successive values,

Summing the sequence of differences to a first sum over a first period,

Summing the sequence of differences to a second sum over a second period longer than the first period,

- Comparison of the first sum and the second sum, and

- Carrying out a classification change depending on the comparison.

2. The method of claim 1, wherein the audio feature is an average or a variance of a level of the audio signal (8).

3. The method according to any one of the preceding claims, wherein the comparison is carried out by means of a quotient of the first sum and the second sum.

A method according to any one of the preceding claims, wherein a temporal succession of values of different audio features is generated and the difference of individual differences is formed by the successive values of similar audio features.

5. The method according to claim 4, wherein the individual differences in the formation of the difference are weighted in the manner of the respective audio recordings.

6. The method according to any one of claims 4 or 5, wherein the values of the various audio features to a feature Vector are summarized and the difference is obtained in the form of a distance between successive feature vectors.

7. The method according to any one of the preceding claims, wherein the classification change is performed depending on a currently selected ^¬ th classification.

A method according to any one of the preceding claims, wherein the first period indicates a duration of 2 seconds to 5 and the second period has a duration of 10 seconds 20 seconds.

9. A method for classifying an audio signal (8) comprising the following steps:

Providing the audio signal (8),

Generating a first value for an audio feature from the audio signal at a first time,

Selecting a classification of the audio signal (8) in dependence on the first value of the audio feature,

Generating a second value for the audio feature from the audio signal (8) at a second time,

Preparation of a classification change by selecting a proposed classification according to the second value of the audio feature,

- generating a time sequence of values of the audio characteristic of the audio signal (8) in an interval between the ers ^¬ th time and the second time,

Formation of a chronological sequence of differences of successive values,

Summing the sequence of differences to a first sum over a first period,

- Comparison of the first sum and the second sum, and - carrying out the classification in accordance with the change before ^¬ impact for a customized classification in dependence on the comparison.

10. The method of claim 9, wherein the classification change is performed depending on the proposal for the adapted Klassifika ^¬ tion.

11. The method according to claim 9 or 10, further developed by the features according to one of claims 1 to 8.

12. Signal processor (20) for classifying an audio signal (8), comprising:

an input interface for receiving an audio signal (8),

- a classification unit (21) for carrying out the Ver ^¬ driving according to one of claims 9 to 11, and

a classification output (22) for outputting the classification as a function of a result of the method applied to the audio signal (8).

13. hearing aid (13), comprising:

a microphone (14) for providing an audio signal (8),

a signal processor (20) according to claim 12,

an audio processor (23) for processing the audio signal (8) according to a processing program depending on the classification of the audio signal (8),

a receiver (17) for outputting the processed audio signal.