EP3386215B1 - Hearing aid and method for operating a hearing aid - Google Patents

Description

The invention relates to a method for operating a hearing device and a hearing device which is set up in particular to carry out the method.

Hearing devices are usually used to output a sound signal to the hearing of the wearer of this hearing device. The output takes place by means of an output transducer, mostly acoustically via airborne sound by means of a loudspeaker (also referred to as "listener" or "receiver"). Such hearing devices are often used as so-called hearing aids (also known as hearing aids for short). For this purpose, the hearing devices normally include an acoustic input transducer (in particular a microphone) and a signal processor which is set up to process the input signal (also: microphone signal) generated by the input transducer from the ambient sound using at least one usually user-specifically stored signal processing algorithm in such a way that a Hearing loss of the wearer of the hearing device is at least partially compensated for. In the case of a hearing aid in particular, the output transducer can also be a so-called bone conduction receiver or a cochlear implant, in addition to a loudspeaker, which are set up for mechanical or electrical coupling of the audio signal into the wearer's hearing. The term hearing devices also includes, in particular, devices such as so-called tinnitus maskers, headsets, headphones and the like.

Modern hearing devices, in particular hearing aids, often include a so-called classifier, which is usually designed as part of the signal processor that executes the respective signal processing algorithm or algorithms. Such a classifier is usually in turn an algorithm which is used to infer an existing hearing situation on the basis of the ambient sound recorded by means of the microphone. On the basis of the recognized hearing situation, the respective signal processing algorithm or algorithms are then usually adapted to the characteristic properties of the present hearing situation. In particular, the hearing device is intended to pass on the information relevant to the user in accordance with the hearing situation. For example, different settings (parameter values of different parameters) of the or one of the signal processing algorithms are required for the clearest possible output of music than for the intelligible output of speech in the case of loud ambient noise. The correspondingly assigned parameters are then changed as a function of the recognized hearing situation.

Usual listening situations are e.g. B. Speech at rest, speech with background noises, listening to music, (driving in) a vehicle. To analyze the ambient sound (specifically the microphone signal) and to identify the respective listening situation, various features (often also referred to as "features") are first derived from the microphone signal (or an input signal formed from it). These features are fed to the classifier, which with the help of analysis models such as B. a so-called "Gaussian mixed mode analysis", a "Hidden Markov model", a neural network or the like outputs probabilities for the presence of certain hearing situations.

A classifier is often "trained" for the respective hearing situation by means of databases in which a large number of different representative audio samples are stored for the respective hearing situations. The disadvantage of this, however, is that in most cases not all combinations of noises that may occur in everyday life can be mapped in such a database. This can therefore lead to misclassification of some listening situations.

EP1858291 A1 describes a method for operating a hearing system which comprises a transmission unit and input / output units linked therewith. A transfer function of the transfer unit describes how audio signals generated by the input unit are processed in order to derive audio signals which are fed to the output unit and which can be set by one or more transfer parameters.

US 2003/0144838 A1 describes a method and a device for identifying an acoustic scene, wherein an acoustic input signal is processed in at least two processing stages so that an extraction phase is provided in at least one of the processing stages, in which characteristic features are extracted from the input signal, and wherein in each processing stage an identification phase is provided in which the extracted characteristic features are classified. According to the classification of the features, class information that characterizes or identifies the acoustic scene is generated in at least one of the processing stages.

WO 2008/084116 A2 describes a method for operating a hearing apparatus comprising an input transducer, an output transducer and a signal processing unit for processing an output signal of the input transducer in order to obtain an input signal for the output transducer by applying a transfer function to the output signal of the input transducer. The method comprises the steps of: extracting features of the output signal of the input transducer, classifying the extracted features by at least two classification experts, weighting the outputs of the at least two classification experts by a weight vector in order to obtain a classification output, setting at least some parameters of the transfer function according to the classification output, Monitoring a user feedback received from the hearing device and updating the weight vector and / or one of the at least two classification experts in accordance with the user feedback.

In EP 2 670 168 A1 Among other things, systems and methods for adaptive hearing aids using several environment detection and classification methods are presented. One aspect of the present topic includes a method of operating a hearing aid for a wearer. Acoustic inputs are received and a variety of acoustic environments determined by parallel signal processing based on the received acoustic inputs. According to various embodiments, an audiological parameter of the hearing aid device is adapted based on the determined plurality of acoustic environments.

The invention is based on the object of making an improved hearing device possible.

This object is achieved according to the invention by a method for operating a hearing device with the features of claim 1. Furthermore, this object is achieved according to the invention by a hearing device with the features of claim 13. Advantageous embodiments and further developments of the invention are in the subclaims and in the following Description set out.

The method according to the invention is used to operate a hearing device which comprises at least one microphone for converting ambient sound into a microphone signal. According to the method, a number of characteristics (also referred to as “features”) are derived from the microphone signal or an input signal formed therefrom. At least three classifiers, which are implemented independently of one another for the analysis of a (preferably permanently) assigned acoustic dimension, are each supplied with a specifically assigned selection from these features. By means of the respective classifier, information is then generated in each case about a characteristic of the acoustic dimension assigned to this classifier. Depending on at least one of the at least three pieces of information about the respective characteristics of the assigned acoustic dimension, at least one signal processing algorithm is then used, which is used to process the microphone signal or the Input signal is processed into an output signal (ie executed), changed.

Changing the signal processing algorithm is understood here and below in particular to mean that at least one parameter contained in the signal processing algorithm is set to a different parameter value as a function of the characteristic of the acoustic dimension or at least one of the acoustic dimensions. In other words, another setting of the signal processing algorithm is "approached" (i.e., effected or made).

The term “acoustic dimension” is understood here and below to mean a group of listening situations that are related due to their specific properties. The hearing situations depicted in such an acoustic dimension are preferably each described by the same features and differ in particular on the basis of the current value of the respective features.

The term "expression" of the respective acoustic dimension is understood here and in the following in particular as to whether (in the sense of a binary distinction) or (in a preferred variant) to what degree (for example, to what percentage) the respective in the respective acoustic dimension the listening situation shown is present. Such a degree or percentage preferably represents a probability value for the presence of the respective hearing situation. For example, in an acoustic dimension directed to the presence of speech, the listening situations "speech in rest", "speech in background noise" or (in particular only) "Background noise" (ie there is no speech), the information about the expression preferably again contains percentages (for example 30% probability for speech in background noise and 70% probability for only background noise).

As described above, the hearing device according to the invention comprises at least one microphone for converting the ambient sound into the microphone signal and a signal processor in which at least the three above-described classifiers are implemented independently of one another for analyzing the respectively (preferably permanently) assigned acoustic dimension. The signal processor is set up to carry out the method according to the invention, preferably automatically. In other words, the signal processor is set up to derive the number of features from the microphone signal or the input signal formed therefrom, to supply each of the three classifiers with a specifically assigned selection from the features, with the aid of the respective classifier information about the expression of the respectively assigned acoustic To generate dimension and, depending on at least one of the three items of information, to change at least one signal processing algorithm (preferably correspondingly assigned to the acoustic dimension) and preferably to apply it to the microphone signal or the input signal.

In a preferred embodiment, the signal processor (also referred to as a signal processing unit) is formed at least in its core by a microcontroller with a processor and a data memory in which the functionality for carrying out the method according to the invention is implemented in the form of operating software ("firmware"), so that the method - possibly in interaction with a user of the hearing device - is carried out automatically when the operating software is executed in the microcontroller. Alternatively, the signal processor is formed by a non-programmable electronic component, e.g. an ASIC, in which the functionality for carrying out the method according to the invention is implemented using circuitry.

The fact that, according to the invention, at least three classifiers are set up and provided for analyzing an associated acoustic dimension and thus in particular for recognizing one hearing situation in each case, advantageously enables at least three hearing situations to be recognized independently of one another. This increases the flexibility of the hearing device in the Recognition of listening situations advantageously increased. The invention is based on the knowledge that at least some listening situations can also be completely independent (ie in particular not influencing one another or only influencing one another to an insignificant extent) and parallel to one another. By means of the method according to the invention and by means of the hearing device according to the invention, the risk can thus be reduced that, at least with regard to the at least three acoustic dimensions analyzed by means of the respectively assigned classifier, mutually exclusive and in particular contradicting classifications (i.e. assessment of the currently existing acoustic situation) can be reduced. comes. In particular, (completely) parallel listening situations can be recognized in a simple manner and taken into account in the change in the signal processing algorithm.

The hearing device according to the invention has the same advantages as the method according to the invention for operating the hearing device.

In a preferred variant of the method, several, ie at least two or more signal processing algorithms are used, in particular in parallel for processing the microphone signal or the input signal. The signal processing algorithms "work" preferably on (at least) one assigned acoustic dimension, ie the signal processing algorithms are used to process (e.g. filtering, amplifying, attenuating) signal components that are relevant for the listening situations contained or mapped in the assigned acoustic dimension . To adapt the signal processing as a function of the characteristics of the respective acoustic dimension, the signal processing algorithms comprise at least one, preferably several parameters, the parameter values of which can be changed. The parameter values can preferably also be changed in several steps (gradually or continuously) depending on the respective probability of the expression. This enables signal processing that is particularly flexible and advantageously adaptable to a large number of gradual differences between a number of listening situations.

According to the method, a different selection from the features is also supplied to at least two of the at least three classifiers. This is understood here and in the following in particular to mean that a different number and / or different features are selected for the respective classifier and supplied to it.

The conjunction “and / or” is to be understood here and below in such a way that the features linked by means of this conjunction can be designed both together and as alternatives to one another.

In a further expedient variant of the method, each of the classifiers is supplied with the correspondingly assigned selection, in particular features that are only relevant for the analysis of the assigned acoustic dimension. In other words, only those features are selected and supplied for each classifier that are actually required to determine the hearing situation mapped in the respective acoustic dimension. In this way, when analyzing the respective acoustic dimension, computational effort and effort in the implementation of the respective classifier can advantageously be saved, since features that are insignificant for the respective acoustic dimension are not taken into account from the outset. The risk of a misclassification due to an erroneous consideration of non-relevant features can thereby advantageously be further reduced.

In an advantageous variant of the method, especially in the event that only the respectively relevant features are used in each classifier, a specific analysis algorithm is used for each of the classifiers to evaluate the (each specifically) supplied features. This, in turn, can also advantageously save computational effort. Furthermore, comparatively complicated algorithms or analysis models such as B. Gaussian mixed modes, neural networks or hidden Markov models, which are used in particular for the analysis of a large number of different, mutually independent features, are omitted. Rather, in particular, each of the classifiers is thus on a specific "problem", ie with regard to its own Analysis algorithm "tailored" (ie adapted or designed) to the acoustic dimension specifically assigned to this classifier. The above-described, comparatively complex analysis models can nevertheless be used for specific acoustic dimensions within the scope of the invention, whereby, due to the orientation of the corresponding classifier to one or a few listening situations included in the specific acoustic dimension, there is an effort involved in implementing such a comparatively complex model can be saved.

In a preferred variant of the method, in particular the dimensions “vehicle”, “music” and “language” are used as the at least three acoustic dimensions. In particular, it is determined within the respective acoustic dimension whether the user of the hearing device is in a vehicle, is actually driving this vehicle, is listening to music or whether speech is present. In the latter case, it is preferably determined within the scope of this acoustic dimension whether speech is present in quiet, speech in background noise or no language and preferably only background noise. These three acoustic dimensions are, in particular, the dimensions that usually occur particularly frequently in the everyday life of a user of the hearing device and are also independent of one another. In an optional development of this variant of the method, a fourth classifier is used to analyze a fourth acoustic dimension, which is in particular the loudness (also: “volume”) of ambient noises (also referred to as “interference noises”). The characteristics of this acoustic dimension extend gradually or continuously over several intermediate stages from very quiet to very loud. In contrast, the information on the characteristics, in particular of the acoustic dimensions vehicle and music, can optionally be "binary", ie it is only recognized whether or not the vehicle is being driven, or whether music is being heard or not. Preferably, however, all information of the other three acoustic dimensions is continuously available as a kind of probability value. This is particularly advantageous because errors in the analysis of the respective acoustic dimension cannot be ruled out, and because this also makes it simple, in contrast to binary information Way "smoother" transitions between different settings can be achieved.

In additional or optionally alternative developments, further classifiers are used in each case for wind and / or reverberation estimation and for the detection of the wearer's own voice.

In an expedient variant of the method, features are derived from the microphone signal or the input signal that are selected from a (in particular non-final) group, which in particular includes the features of signal level, 4 Hz envelope modulation, onset content, level of background noise (also called “Noise Floor Level” denotes, optionally at a predetermined frequency), the spectral focus of the background noise, stationarity (in particular at a predetermined frequency), tonality and wind activity.

In a further expedient variant of the method, at least the characteristics level of the background noise, spectral focus of the background noise and stationarity (and optionally also the characteristic of wind activity) are assigned to the acoustic dimension vehicle. The characteristics of onset content, tonality and level of the background noise are preferably assigned to the acoustic dimension of music. The characteristics of onset content and 4 Hz envelope modulation in particular are assigned to the acoustic dimension of language. The loudness dimension of the ambient noise, which may be present, is assigned in particular the characteristics of the level of the background noise, the signal level and the spectral focus of the background noise.

In a further expedient variant of the method, a specifically assigned temporal stabilization is taken into account for each classifier. In particular, with some of the classifiers it is preferably assumed that this state is present in the past (for example, in a previous time segment of predetermined duration) (ie in particular with a certain characteristic of the acoustic dimension) (the expression) is then still present with a high degree of probability at the current point in time. For this purpose, for example, a sliding mean value is formed over (in particular a predetermined number of) previous time segments. Alternatively, a type of "dead time element" can also be provided, by means of which, in a subsequent time segment, the probability is increased that the expression present in the previous time segment is still present. For example, if driving in the vehicle was recognized in the previous five minutes, it is assumed that this situation continues to exist. Comparatively “strong” stabilizations are preferably used for the dimensions vehicle and music, ie only comparatively slow or rare changes in the correspondingly assigned hearing situations are assumed. For the language dimension, on the other hand, no stabilization or only a “weak” stabilization is expediently undertaken, since rapid and / or frequent changes in the listening situation are assumed here. Speech situations can often only last a few seconds (e.g. about 5 seconds) or a few minutes, whereas driving in the vehicle usually lasts for several minutes (e.g. more than 3 to 30 minutes or even hours). A further optional variant for stabilization can also take place via a counting principle in which a counter ("counter") is incremented with a comparatively fast (e.g. 100 milliseconds to a few seconds) detection cycle and the "detection" is only activated when a limit value is exceeded for this counter "the respective hearing situation is triggered. This is useful, for example, as short-term stabilization in the case of a common classifier for "all" listening situations. As a modification to stabilization in the present case, it is, for example, conceivable to assign a separate limit value to each listening situation and to reduce this value, in particular for the listening situation "driving in the vehicle" and / or "listening to music", if the respective limit value has already been set for a predetermined preceding period of time Listening situation has been recognized.

In a further expedient variant of the method, the respective signal processing algorithm or the respective signal processing algorithm is or are dependent on at least two of the at least three pieces of information about the characteristics of the respectively assigned acoustic Dimension adjusted. The information from several classifiers is therefore taken into account in at least one signal processing algorithm.

In an expedient variant of the method, the respective information from the individual classifiers is in particular first fed to a merger element for a common evaluation (“merged”). On the basis of this joint evaluation of all information, in particular overall information about the present listening situations is created. A dominant hearing situation is preferably determined in this case - in particular on the basis of the degree of expression reflecting the probability. The respective signal processing algorithm or algorithms are adapted to this dominant hearing situation. A hearing situation (namely the dominant one) is optionally prioritized by changing the respective signal processing algorithm only as a function of the dominant hearing situation, while other signal processing algorithms and / or the parameters dependent on other hearing situations remain unchanged or to a parameter value that does not have any Has an impact on signal processing.

In a further development of the method variant described above, based on the joint evaluation of all information, in particular a hearing situation referred to as a sub-situation is determined, which has a lower dominance compared to the dominant hearing situation. This or the respective sub-situation is additionally taken into account in the aforementioned adaptation of the respective signal processing algorithm or algorithms to the dominant hearing situation and / or to adapt a signal processing algorithm specifically assigned to the acoustic dimension of this sub-situation. In particular, this sub-situation leads to a smaller change in the parameter or parameters assigned in each case compared to the dominant hearing situation. In the event that, for example, speech is determined in background noise as the dominant listening situation and music is used as the sub-situation, one or more parameters of a signal processing algorithm, which is used to ensure the clearest possible speech intelligibility in the case of background noise, are changed comparatively strongly in order to achieve the highest possible speech intelligibility. But since there is also music, parameters that serve to attenuate ambient noise are set less strongly (than if only background noise is present) so that the tones of the music are not completely attenuated. A signal processing algorithm (in particular additional) serving for clear sound reproduction of music is also set less strongly than with music as the dominant listening situation (but stronger than with no music) so as not to cover up the speech components. Thus, in particular because of the mutually independent detection of different hearing situations and because of the finer adaptation of the signal processing algorithms made possible by this, a particularly precise adaptation of the signal processing of the hearing device to the actually present hearing situation can take place.

As already described above, the parallel presence of several hearing situations is preferably taken into account in at least one of the possibly several signal processing algorithms.

In an alternative variant of the method, the or preferably each signal processing algorithm is assigned to at least one of the classifiers. In this case, preferably at least one parameter of each signal processing algorithm is changed (in particular directly) as a function of the information output by the respective classifier about the characteristic of the assigned acoustic dimension. This parameter or its parameter value is preferably designed as a function of the respective information. In this way, the information about the characteristics of the respective acoustic dimension is used, in particular, directly for adapting the signal processing. In other words, each classifier "controls" at least one parameter of at least one signal processing algorithm. A joint evaluation of all information can be omitted here. In particular, in this case a particularly large amount of information about the distribution of the mutually independent hearing situations in the currently available "image" described by the ambient sound is taken into account, so that in turn a particularly fine adjustment of the signal processing is promoted. In particular, completely parallel listening situations - for example 100% speech in background noise 100% driving in the vehicle, or 100% music with 100% driving in the vehicle - can be taken into account in a simple manner and with little loss of information.

In a further expedient variant of the method, at least one of the classifiers is supplied with status information that is generated independently of the microphone signal or the input signal. This status information is also taken into account, in particular, for evaluating the respective acoustic dimension. For example, this involves movement and / or location information that is used, for example, to evaluate the acoustic dimension of the vehicle. This movement and / or location information is generated, for example, with an acceleration or (global) position sensor arranged in the hearing device itself or in a system connected to it for signal transmission purposes (e.g. a smartphone). For example, using an existing movement speed (having a predetermined value) when evaluating the acoustic dimension vehicle, the probability of the presence of the hearing situation driving in the vehicle can be increased in a simple manner in addition to the acoustic evaluation. One speaks here of "augmenting" a classifier.

Fig. 1

in einer schematischen Übersichtsdarstellung eine Hörvorrichtung,

Fig. 2

in einem schematischen Blockschaltbild einen Signallaufplan der Hörvorrichtung gemäß Fig. 1,

Fig. 3

in einem schematischen Ablaufplan ein Verfahren zum Betrieb der Hörvorrichtung gemäß Fig. 1, und

Fig. 4

in Ansicht gemäß Fig. 2 ein alternatives Ausführungsbeispiel des Signallaufplans.

Exemplary embodiments of the invention are explained in more detail below with reference to a drawing. Show in it:

Fig. 1

in a schematic overview representation of a hearing device,

Fig. 2

in a schematic block diagram, a signal flow diagram of the hearing device according to FIG Fig. 1 ,

Fig. 3

in a schematic flow chart a method for operating the hearing device according to FIG Fig. 1 , and

Fig. 4

in view according to Fig. 2 an alternative embodiment of the signal flow diagram.

Corresponding parts and sizes are always provided with the same reference symbols in all figures.

In Figure 1 a hearing aid device, referred to as "hearing device 1" for short, is shown as a hearing device. The hearing aid 1 comprises, as electrical components housed in a housing 2, two microphones 3, a signal processor 4 and a loudspeaker 5 ) or as a secondary cell (ie as a rechargeable battery). When the hearing device 1 is in operation, ambient sound is recorded by means of the microphones 3 and a microphone _{signal S M is} generated therefrom. These two microphone signals S _M are fed to the signal processor 4, which, while processing four signal processing algorithms A ₁ , A ₂ , A ₃ and A _4, generates an output _{signal S A} from these microphone signals S _M and sends this to a loudspeaker 5, which is an output transducer, issues. The loudspeaker 5 converts the output signal S _A into airborne sound, which is transmitted to the hearing of a user or wearer (short: hearing aid wearer) of the hearing aid via a sound tube 7 connected to the housing 2 and an earpiece 8 connected to it at the end (when the hearing aid 1 is properly worn) 1 is issued.

In order to recognize different hearing situations and to subsequently adapt the signal processing, the hearing device 1, specifically its signal processor 4, is set up to use a method described below on the basis of FIG Figure 2 and Figure 3 to carry out the procedure described in more detail automatically. As in Figure 2 Shown in more detail, the hearing aid 1, specifically its signal processor 4, comprises at least three classifiers Ks, K _M and K _F. These three classifiers K _S , K _M and K _F are each set up and designed to analyze a specifically assigned acoustic dimension. The classifier Ks is specifically designed to evaluate the acoustic dimension “language”, ie whether language, language is present in background noise or just background noise. The classifier K _M is specifically designed to evaluate the acoustic dimension “music”, ie whether the ambient sound is dominated by music. The classifier K _F is specifically designed to evaluate the acoustic dimension “vehicle”, ie to determine whether the hearing aid wearer is driving in a vehicle. The signal processor 4 further comprises a feature analysis module 10 (also referred to as a "feature extraction module") which is set up to derive a number of (signal) features from the _{microphone signals S M} , specifically from an input signal S _{E formed from these microphone signals S M.} The classifiers Ks, K _M and K _F are each supplied with a different and specifically assigned selection from these features. On the basis of these specifically supplied features, the respective classifier Ks, K _M or K _{F determines} a characteristic of the respective assigned acoustic dimension, ie the degree to which a hearing situation specifically assigned to the acoustic dimension is present, and outputs this characteristic as the respective information.

How to get concrete Figure 3 It can be seen, in a first method step 20, the microphone _{signals S M are} generated from the detected ambient sound and are combined by the signal processor 4 to form the input signal S _E (specifically mixed to form a directional microphone signal). In a second method step 30, the input signal S _{E formed from the microphone signals S M is} fed to the feature analysis module 10, and the number of features is derived therefrom. Specific (but not conclusive) characteristics are the level of a background noise (characteristic "M _P "), a spectral focus of the background noise (characteristic "M _Z "), a stationarity of the signal (characteristic "M _M "), wind activity ( Feature "M _W "), an onset content of the signal (feature "M _O "), a tonality (feature "M _T ") and a 4-Hertz envelope modulation (feature "M _E ") are determined. In a method step 40, the features M _E and Mo are fed to the classifier Ks for analyzing the acoustic dimension of speech. The features M _O , M _T and M _P are fed to the classifier K _M for analyzing the acoustic dimension of music. _{The characteristics M P} , M _W , M _Z and M _{M are} fed to the classifier K _F to analyze the acoustic dimension of driving in the vehicle. The classifiers K _S , K _M and K _F then use specifically adapted analysis algorithms to determine the extent to which, ie to what degree, the respective acoustic dimension is pronounced on the basis of the features supplied in each case. Specifically, the classifier Ks is used to determine the probability with which speech is present at rest, speech in background noise or only background noise. The classifier K _M is used to determine with what probability there is music. The classifier K _F is used to determine the probability with which the hearing aid wearer is driving or not in a vehicle.

In an alternative exemplary embodiment, it is only determined in a “binary” manner whether speech, possibly in background noise or only background noise, or whether there is music or driving in the vehicle or not.

The respective characteristics of the acoustic dimensions are output to a fusion module 60 in a method step 50 (see FIG Figure 2 ) by bringing the respective information together and comparing it with one another. In addition, a decision is made in the fusion module 60 as to which dimension, specifically which hearing situation depicted therein, is currently to be regarded as dominant and which hearing situations are currently of subordinate importance or can be completely excluded. The fusion module then changes a number of the parameters relating to the dominant and the less relevant hearing situations with a number of the stored signal processing algorithms A ₁ to A ₄ , so that the signal processing is primarily adapted to the dominant hearing situation and to a lesser extent to the less relevant hearing situation. Each of the signal processing algorithms A ₁ to A ₄ is in each case adapted to the presence of a hearing situation, possibly also parallel to other hearing situations.

The classifier K _F includes a stabilization over time in a manner not shown in detail. This is geared in particular to the fact that a journey in the vehicle usually lasts a longer time, and thus in the event that driving in the vehicle has already been recognized in previous time segments, for example from 30 seconds to five minutes each, and on the assumption that the driving situation in the vehicle is still ongoing, the probability that this hearing situation is present is already increased in advance. The same is set up and provided in the classifier K _M.

In an alternative embodiment according to Figure 4 The fusion module 60 is missing in the signal flow diagram shown. Each of the classifiers Ks, K _M and K _F is assigned at least one of the signal processing algorithms A ₁ , A ₂ , A ₃ and A _{4 in} such a way that several in the respective signal processing algorithm A ₁ , A ₂ , A ₃ or A ₄ contained parameters are designed to be changeable as a function of the characteristics of the respective acoustic dimension. This means that on the basis of the respective information about the respective characteristic, at least one parameter is changed directly - that is, without any intermediate merging. Specifically, in the exemplary embodiment shown, the signal processing _{algorithm A 1 is} only dependent on the information from the classifier Ks. In contrast, the information from all classifiers K _S , K _M and K _F flow into the signal processing _{algorithm A 3} and lead to a change in several parameters there.

The subject matter of the invention is not limited to the exemplary embodiments described above. Rather, further embodiments of the invention can be derived from the above description by a person skilled in the art. In particular, the individual features of the invention described on the basis of the various exemplary embodiments and their design variants can also be combined with one another in other ways. For example, the hearing aid 1 can also be designed as an in-the-ear hearing aid instead of the behind-the-ear hearing aid shown. The subject matter of the invention is defined in the following claims.

List of reference symbols

1

Hearing aid

2

casing

3

microphone

4th

Signal processor

5

speaker

6th

battery

7th

Sound tube

8th

Earpiece

10

Feature analysis module

20th

Process step

30th

Process step

40

Process step

50

Process step

60

Fusion module

A1-A4

Signal processing algorithm

KS, KM, KF

Classifier

ME, MO, MT, MP, MW, MZ, MM

characteristic

SA

Output signal

SE

Input signal

SM

Microphone signal

Claims (13)

1. Method for operating a hearing device (1), which comprises at least one microphone (3) for converting ambient sound into a microphone signal (S_M), wherein according to the method

   - a plurality of features (M_E, M_O, M_T, M_P, M_W, M_Z, M_M) are derived from the microphone signal (S_M) or from an input signal (S_E) formed therefrom,

   - at least three classifiers (K_S, K_M, K_F), which are implemented independently of each other for respectively analyzing an assigned acoustic dimension - i.e. a group of hearing situations, which are related on the basis of their specific properties - are each supplied with a specifically assigned selection from these features (M_E, M_O, M_T, M_P, M_W, M_Z, M_M), wherein at least two of the at least three classifiers (K_S, K_M, K_F) each are supplied with a different selection from the features (M_E, M_O, M_T, M_P, M_W, M_Z, M_M),

   - by means of the respective classifier (K_S, K_M, K_F), in each case an information is generated about a characteristic of the acoustic dimension assigned to this classifier (K_S, K_M, K_F), and

   - at least one signal processing algorithm (A₁, A₂, A₃, A₄), which is executed in order to process the microphone signal (S_M) or the input signal (S_E) into an output signal (S_A), is modified as a function of at least one of the at least three pieces of information about the respective characteristic of the assigned acoustic dimension.
2. Method according to claim 1,
   wherein only features (M_E, M_O, M_T, M_P, M_W, M_Z, M_M) relevant for analyzing the respectively assigned acoustic dimension are supplied to each of the classifiers (K_S, K_M, K_F) with the correspondingly assigned selection.
3. Method according to one of claims 1 to 2,
   wherein for each of the classifiers (K_S, K_M, K_F) a specific analysis algorithm is used for evaluating the respective supplied features (M_E, Mo, M_T, M_P, M_W, M_Z, MM).
4. Method according to one of claims 1 to 3,
   wherein vehicle, music and speech are used as the at least three acoustic dimensions.
5. Method according to one of claims 1 to 4,
   wherein characteristics selected from signal level, 4-hertz envelope modulation (M_E), onset content (M_O), level of a background noise (M_P), spectral center of gravity of the background noise (M_Z), stationarity (M_M), tonality (M_T), wind activity (M_W) are derived from the microphone signal (S_M) or respectively the input signal (S_E).
6. Method according to one of claims 4 and 5,
   wherein at least the features level of background noise (M_P), spectral center of gravity of background noise (M_Z) and stationarity (M_M) are assigned to the acoustic dimension vehicle, wherein the features onset content (M_O), tonality (M_T) and level of background noise (M_P) are assigned to the acoustic dimension music, and wherein the features onset content (M_O) and 4-hertz envelope modulation (M_E) are assigned to the acoustic dimension speech.
7. Method according to one of claims 1 to 6,
   wherein for each classifier (K_S, K_M, K_F) a specifically assigned temporal stabilization is taken into consideration.
8. Method according to one of claims 1 to 7,
   wherein the or the respective signal processing algorithm (A₁, A₂, A₃, A₄) is modified as a function of at least two of the at least three pieces of information about the manifestation of the respectively assigned acoustic dimension.
9. Method according to one of claims 1 to 8,
   wherein the information of the respective classifiers (K_S, K_M, K_F) are supplied to a joint evaluation, wherein a dominant hearing situation is determined on the basis of this joint evaluation, and wherein the or the respective signal processing algorithm (A₁, A₂, A₃, A₄) is adapted to this dominant hearing situation.
10. Method according to claim 9,
    wherein at least one subsituation with lower dominance in comparison to the dominant listening situation is determined, and wherein this or the respective subsituation is taken into account in the modification of the signal processing algorithm (A₁, A₂, A₃, A₄) or at least one of the signal processing algorithms (A₁, A₂, A₃, A₄).
11. Method according to one of claims 1 to 7,
    wherein each signal processing algorithm (A₁, A₂, A₃, A₄) is assigned to at least one of the classifiers (K_S, K_M, K_F), and wherein at least one parameter of each signal processing algorithm (A₁, A₂, A₃, A₄) is modified as a function of the information about the manifestation of the corresponding acoustic dimension output by the assigned classifier (K_S, K_M, K_F).
12. Method according to one of claims 1 to 11,
    wherein at least one of the classifiers (K_S, K_M, K_F) is supplied with a status information generated independently of the microphone signal (S_M) or the input signal (S_E), which status information is taken into account in addition to the evaluation of the respective acoustic dimension.
13. Hearing device (1),

    - with at least one microphone (3) for the conversion for converting ambient sound into a microphone signal (S_M), and

    - with a signal processor (4), in which at least three classifiers (Ks, K_M, K_F) are implemented independently of each other for respectively analyzing an assigned acoustic dimension - i.e. a group of hearing situations, which are related on the basis of their specific properties - and wherein the signal processor (4) is configured to perform the method according to one of claims 1 to 12.

Images