Method for signal processing of an external audio signal in a hearing instrument

Abstract

The invention relates to a method for signal processing of an external audio signal (Ext) in a hearing instrument (1), wherein a data signal (DS) containing the external audio signal (Ext) is received by a communication device (5) of the hearing instrument (1), wherein the external audio signal (Ext) is examined with regard to a first set of acoustic features (Fa-n), wherein a first classifier (DNN1) determines a first probability value (w1a-c) for a first plurality of first classes (C1a-c) of audio content based on the acoustic features (Fa-n) of the first set as input variables, which first probability value indicates a probability for the presence of the respective first class (C1a-c) of audio content in the external audio signal (Ext), and wherein at least one parameter of a signal processing unit (7) of the hearing instrument (1), which generates a playback signal (A1) based on the external audio signal (Ext),is set based on at least one of the first probability values ​​(w1a-c).

Description

The invention relates to a method for signal processing of an external audio signal in a hearing instrument, wherein a data signal containing the external audio signal is received by a communication device of the hearing instrument, and wherein signal processing of the hearing instrument generates a reproduction signal on the basis of the external audio signal.The term "hearing instrument" is usually understood to mean devices which are used for outputting sound signals to the hearing or, more generally, to the hearing center of a user of the corresponding device. In particular, this term includes hearing aids. Hearing aids serve to at least partially compensate for a hearing loss resulting from this hearing reduction. Hearing aids usually have at least one electroacoustic input transducer, usually in the form of a microphone, for detecting acoustic (ambient) noise and converting it into an electrical input signal. Furthermore, such hearing devices regularly have a signal processing device which is configured to analyze the input signal or signals for interference components (e.g. noise, ambient noise and the like), to filter and / or attenuate these interference components and to amplify the remaining signal components as a useful signal (such as in particular speech and / or music).For outputting the input signal processed in this way to the hearing, hearing aids usually comprise an electroacoustic output transducer, for example in the form of a loudspeaker (also referred to as a listener or "receiver"), by means of which the processed input signal is converted into an output sound signal and output to the hearing aid wearer. Alternatively, hearing aids have a cochlea or bone conduction earphone for outputting an output signal in electrical or mechanical form to the hearing.However, the term "hearing instrument" also includes other devices for sound output, such as headsets ("earphones"), wireless earphones with and without active noise suppression, so-called "hearables" and the like.Hearing instruments in general and also hearing aids in the narrower sense are often configured to generate a reproduction signal for generating said output sound signal not only on the basis of recording and (optionally user-specific) processing of the ambient sound, but also to receive external audio signals as data via suitable antennas and / or downlink devices, for example in the form of streaming signals, and to process the external audio signal into the reproduction signal. Not least in the case of hearing devices for providing a hearing deficiency, however, it is advantageous to tune the aforementioned processing of the external audio signal (i.e., for example, of the data of the streaming signal) to the playback signal to a content of the external audio signal, i.e., to choose a different signal processing for a podcast or a hearing book than for music content (and different settings of the signal processing may also be advantageous for these depending on the direction of music).Such information can often be contained in metadata of the data packets in which the external audio signal is transmitted to the hearing instrument. This type of information is, however, not available for all external audio signals.The invention is therefore based on the object of specifying the most advantageous possible processing of an external audio signal, which is transmitted to a hearing instrument, to form a reproduction signal in the hearing instrument.The object mentioned is achieved according to the invention by a method for signal processing of an external audio signal in a hearing instrument, wherein a data signal which contains the external audio signal is received by a communication device of the hearing instrument, wherein the external audio signal is examined with respect to a first set of acoustic features, wherein a first classifier determines a first probability value for a first plurality of first classes of audio contents on the basis of the acoustic features of the first set as input variables, which probability value indicates a probability for a presence of the respective first class of audio contents in the external audio signal, and wherein at least one parameter of signal processing of the hearing instrument, which generates a reproduction signal on the basis of the external audio signal, is set on the basis of at least one of the first probability values, and preferably the reproduction signal is converted into an output sound signal by an output converter of the hearing instrument. Advantageous and partly per se inventive embodiments are the subject matter of the dependent claims and of the following description.In this case, a hearing instrument generally comprises any device which is configured to generate an electrical input signal from an ambient sound by means of at least one, in particular acousto-electrical, input transducer, process said input signal into a reproduction signal by means of a gain and / or compression, in particular specific to the frequency band, and generate a sound signal from the reproduction signal and supply it to an ear of a wearer of this device, in particular by means of an electro-acoustic output transducer (for example a loudspeaker, a so-called balanced metal case receiver, but also a bone guide earphone). Thus, a hearing instrument includes in particular a headset (e.g. as "earbud"), a headset, smart glasses with loudspeakers, etc., which are equipped with a corresponding input transducer. However, a hearing device in the narrower sense is also included as a hearing instrument, that is to say a device for supplying a hearing weakness of the wearer, in which device, during the processing of the input signal to form the reproduction signal, the former is amplified and / or compressed in particular as a function of the frequency band, in order to at least partially compensate for the hearing weakness of the wearer in a user-specific manner by means of an output sound signal generated from the output signal.In particular, the hearing instrument can also be designed as a binaural hearing system having a first local device and a second local device, wherein in this case the hearing instrument can preferably also have (at least) one further input transducer, wherein the first input transducer is arranged in the first local device, and the further input transducer is arranged in the second local device. The applicability of the described method is generally independent thereof.A data signal includes any analog or digital signal which is transmitted to the hearing instrument by a transmitter device. The transmitting device is in particular part of a superordinate device, e.g. a television set or a speech system or a smartphone or the like. The data signal can contain its information in analog form (i.e. e.g. via continuous frequency and / or phase and / or amplitude modulation) or in digital form as an electromagnetic signal.An external audio signal includes any analog or digital audio signal which is transmitted by the transmitter device (in the present case as information content of the data signal) to the hearing instrument, wherein the external audio signal can be transmitted directly (i.e. the data signal is formed directly by the streamed external audio signal), or the data signal can also contain additional information, for example with regard to the transmission protocol used, or also redundant data packets to increase the robustness of the transmission.In this case, the inclusion of the external audio signal in the data signal includes, in particular, the acoustic information content of the external audio signal being contained in the data signal, and the external audio signal being able to be obtained from the data signal, in particular by means of a suitable decoding method. In other words, the data signal can preferably be understood as a purely information-related signal (for example a sequence of data packets which in turn contain different bits), while the external audio signal is already present as "playable" as an acoustic signal (in particular in one of the usual audio data formats such as e.g. wav, mp3, aac etc.).In this case, the communication device includes in particular any device of the hearing instrument which is provided and configured for receiving the data signal containing the external audio signal, that is to say in particular an antenna.Acoustic features are understood here to mean, in particular, time-dependent acoustic variables which characterize the external audio signal with regard, in particular, to its spectral properties and / or its dynamic and / or spectral time variation and / or with regard to harmonic / tonal components. The advantage of using acoustic features which are determined from the external audio signal instead of using the signal components of the external audio signal itself directly as an input variable of the first classifier is that in this case the acoustic features already provide prefiltering of the information acoustically relevant for the classification, whereby in each case a comparatively manageable set of features may already be sufficient for the classification, while direct signal components would require a significantly more complex model for the classification. In particular, this facilitates implementation of the first classifier by a corresponding artificial neural network (deep neural network, DNN).A classifier includes, in particular, a device or a program element which is(s) configured to determine the respective probability of a presence on the basis of the first acoustic features on input variables for the predefined plurality of first classes of audio content. The first classifier can be implemented as an independent device, for example, via an ASIC or another hard-wired circuit, or also as a program element on a signal processor configured for this purpose. The first classifier is preferably implemented as a DNN.The first classes of audio contents are preferably given by possible groups of acoustically similar contents, i.e. for example "speech", "music", but optionally also "noise", etc. A probability value for a presence of the respective first class of audio contents is in this case related in particular to the probability with which the audio contents on which the respective class is based is contained in the external audio signal "per se". Preferably, the first probability values each assume values from zero to one.The setting of a parameter of a signal processing of the hearing instrument, which generates the reproduction signal on the basis of the external audio signal, here comprises in particular that the signal processing of the hearing instrument generates the reproduction signal, which is converted by the output converter into an output sound signal and is thereby reproduced, and that during this signal processing the external audio signal enters the reproduction signal at least in parts (in particular in spectral components and in particular with at least partial admixture of signal components of a microphone signal), wherein the signal processing can still process the signal components of the external audio signal, that is to say for example via a frequency-band-wise increase or decrease and / or a frequency-band-wise dynamic compression, an additional noise suppression, etc. The signal processing controlling this processing of the external audio signal by the signal processing is performed on the basis of the output signal, at least one parameter (such as gains and / or compression characteristics in one or more frequency bands) is then adjusted based on at least one of the first probabilities.As a result, processing of the external audio signal can be tuned to the type of its content, wherein one does not rely on the presence of additional information by metadata. The described classification by means of the first classifier allows a logical grouping of different contents, which reduces the recognition of the (audio) content of the external audio signal to a recognition of the membership in a given class (instead of applying an arbitrarily complex signal recognition to the external audio signal), and thereby substantially simplifies it. In addition, said classification on the basis of the acoustic features can be implemented efficiently in a DNN, which in turn saves computing power of the processing hardware in the hearing instrument (e.g. signal processor and / or ASIC) and thereby battery power.Preferably, each of the first probability values is determined in such a way that the associated probability for a presence of the respective first class of audio contents in the external audio signal is specified independently of the respective other first probability values. In particular, this means here that the first probability values are independent of one another in the sense that, in the case of a change of only a specific audio content (e.g. a decrease in "speech"), the probability values of other classes (e.g. "noise" or "music") do not necessarily have to change. In particular, this means that the probability values of the first classes do not necessarily have to add to one, since a probability value does not indicate or estimate in the present case which audio content or which class is the predominant one in the external audio signal (that is to say, for example, "of all classes would be the most probable"), but rather whether and to what extent the relevant audio content is present in the external audio signal.Advantageously, the acoustic features of the first set are selected from the following features: center-of-gravity frequency of a total signal and / or a noise background and / or a band-limited signal, variance of the center-of-gravity frequency, modulation depth at at least one given modulation frequency (preferably at a modulation frequency from an interval of 4 Hz±1 Hz), modulation phase difference of at least two different frequency bands, stationarity, signal level, noise level for at least one given frequency range (preferably for a high-frequency and / or a medium-frequency frequency range of the audible spectrum), autocorrelation value for at least one given time delay (for example for a given time delay from an interval of 3 ms±1 ms and / or a given time delay from an interval of 8 ms±2 ms), beat or accent detection, onset detection. The beat or accent recognition can be designed here in particular for recognizing a rhythm ("beat") in the musical sense. The acoustic features mentioned provide a particularly good trade-off between a simple elevation and a high content of information with respect to the audio contents to be classified.Expediently, at least the classes "music" and "speech" (or "speech") are included as first classes, wherein the associated first probability values preferably each indicate the probability for a presence of music or speech in the external audio signal.It has proven to be further advantageous if, on the basis of at least some of the acoustic features of the first set and on the basis of at least one of the first probability values, a second classifier for a second plurality of second classes of audio contents, which are different from the audio contents of the first plurality of first classes, respectively determines a second probability value, which indicates a probability for a presence of the respective second class of audio contents in the external audio signal, and wherein the at least one parameter of the signal processing of the hearing instrument and / or a further parameter of the signal processing of the hearing instrument is set on the basis of at least one of the second probability values. In particular, in addition to the acoustic features of the first set, the second classifier can also use further acoustic features as input variables which are not contained in the first set, and / or neglect individual acoustic features of the first set (and thus not completely identical input variables of the first classifier are used). For the second classifier, which is preferably implemented by a DNN, in particular the above explanations apply to the first classifier, mutatismutanidis.The second plurality of second classes of audio content is different from the audio content of the first plurality of first classes of audio content (i.e., the audio content is not coincident) and preferably completely different (i.e., any class is either a first or a second class), but the second classes may preferably be configured as "sub-classes" of a respective first class (e.g., in terms of the type of "speech" such as "discussion", "single speaker / podcast", "auditory book", or the like, in terms of the type of "music" such as "classic music / orchester", "pop / rock", "electro / techno", or also "music" with gesang, etc.). The at least one parameter is thus set, for example, as a function of a first probability for the first class "speech", and moreover, optionally as a function of a second probability for the second class "classic music / orchester". In particular, the second probabilities can also be used to set other parameters of the signal processing than the first probabilities.Preferably, the at least one of the first probability values is used as an input variable for the second classifier, and / or the second classifier is preferably activated on the basis of the at least one of the first probability values. In the first case, this means in particular that the second classifier changes or adapts its mode of operation to audio contents as a function of said first probability value of the respective first class, and outputs, for example, for a high first probability value with given other input variables, other second probability values for the second classes than for a low first probability value (and otherwise identical input variables). In this case, the first probability value used as input variable of the second classifier is preferably greater than zero and less than one.In the second case mentioned, this means in particular that the first probability value used is used as a binary trigger. If, for example, the first probability value is above an activation limit value, the second classifier is activated. This may be useful in that the above-mentioned conditions are:. Determination of "sub-classes" of audio contents is only meaningful or of interest from a minimum portion for a relevant class, i.e. for example in the case of a low first probability value of the first class "music" as possible second classes, the types of music are on the one hand not highly relevant for the signal processing and on the other hand a classification carried out in this respect may possibly be prone to errors.Preferably, on the basis of the one parameter of the signal processing of the hearing instrument, in particular source-based spectral shaping is controlled in order to form the reproduction signal. This allows specific frequency ranges to be highlighted in a targeted manner, which have particular relevance for the respective audio content (e.g. in the case of "speech", the frequency ranges of formants for vowel recognition or of sisils or plosives; in the case of "music", fundamentally different frequency shaping than in the case of "speech").It has proven to be further advantageous if temporal smoothing is applied to at least one first probability value, and a time constant of said smoothing is set as a function of the same first probability value at a preceding point in time. This can mean, in particular, that such a temporal smoothing of the first probability (via a comparison with corresponding limit values for the application of the smoothing) is "conducted" in a specific direction, that is to say a type of "bias" is applied as a function of the preceding values. Thus, for example, a time constant for an "attack" of the smoothing can be chosen to be shorter and a time constant for a "release" of the smoothing can be chosen to be longer, the longer the second probability value exceeds a predefined limit value of, for example. 0.5 or 0.6, since even a brief drop in the (unsmoothed) first probability value is then not significantly significant (for example, for the first class "music", the first probability value of this class is largely retained even in pauses between individual song). In particular, the method can also be carried out in this way for the second probability values.In a further advantageous embodiment, a binaural hearing system with two local hearing devices is used as hearing instrument, wherein a provisional first probability value is determined in each of the two local hearing devices for each of the first classes, and wherein the first probability value is determined for at least some of the first classes on the basis of the two associated provisional first probability values of the two local hearing devices. In other words, each of the two local hearing devices of the binaural hearing system carries out the method described above on the basis of only the local input signal (or the local input signals) initially independently of the respective other local hearing device, wherein the preliminary first probability values are determined on each side. These are then transmitted to the respective other local hearing device, so that the first probability value can be determined from these on both sides, preferably according to the same rule, for example by averaging or by a maximum or minimum. In particular, the method can also be carried out in this way for the second probability values.The invention further provides a hearing instrument, comprising: a communication device for receiving a data signal which contains an external audio signal, and signal processing means which are configured to implement at least one first classifier, wherein the hearing instrument is configured to carry out the method described above. The hearing instrument is in this case equipped in particular with corresponding signal processors and / or ASICs in order to implement the first classifier.The hearing instrument according to the invention shares the advantages of the method according to the invention. The advantages specified for the method and for its developments can be transferred analogously to the hearing instrument.An exemplary embodiment of the invention is explained in more detail below with reference to drawings. Here, in each case, diagrammatically show: FIG. 1 is a block diagram of a hearing instrument, FIG. 2 is a block diagram of a signal flow for classifying an audio content of an external audio signal in the hearing instrument according to FIG. 1 ; and FIG. 3 is a block diagram of a binaural hearing system as a hearing instrument alternative to FIG. 1.Mutually corresponding parts and sizes are each provided with the same reference numerals in all figures.FIG. 1 schematically shows a block diagram of a hearing instrument 1, which is provided as a hearing device of the BTE design. However, the following statements are also valid for other types of hearing instruments, in particular for those which are not primarily provided for supplying a hearing deficiency.The hearing instrument 1 has a first input transducer M 1 and a second input transducer M 2. The first input transducer M 1 and the second input transducer M 2 are each provided by corresponding microphones. The first input transducer M 1 is configured to generate a first input signal E 1 from an ambient sound 2 during operation of the hearing instrument 1. Accordingly, the second input transducer M 2 is configured to generate a second input signal E 2 from the ambient sound 2 during operation of the hearing instrument 1. The first input signal E 1 and the second input signal E 2 are fed to a signal processing unit 4 in which both input signals E 1, E 2 are processed to form a reproduction signal A 1, and in the process are amplified and / or compressed in particular in a frequency band-specific manner, wherein the said signal processing of the two input signals E 1, E 2 to form the reproduction signal A 1 takes place in particular in a direction-dependent manner, i.e. contributions of individual sound sources from different spatial directions can be amplified to different extents in ambient sound. Moreover, said signal processing can be carried out in particular according to the audiological requirements of a wearer of the hearing instrument 1.The hearing instrument 1 further comprises a communication device 5 which is provided and configured to receive data signals. The data signals can be encoded in particular in electromagnetic waves or transmitted to the hearing instrument 1 by means of magnetic induction.In the present case, the communication device receives a data signal DS, in the data of which an external audio signal Ext is encoded as payload, among other things. Unlike in the schematic illustration of FIG. 1, the data signal DS can be supplied to the signal processing unit and the external audio signal Ext can be decoded there from the data signal DS. However, the external audio signal Ext can also already be decoded by the communication device 5 (and then supplied to the signal processing unit 4) if the communication device 5 is correspondingly configured for this purpose. The external audio signal Ext can be processed into the reproduction signal 1 by the signal processing unit 4 by mixing the external audio signal, possibly after a frequency band-wise processing of its signal components, with the reproduction signal 1 while maintaining (and optionally attenuating) the signal components of the input signals E 1, E 2. In particular, the external audio signal Ext also enters the playback signal 1 by muting the input signals E1, E2.The hearing instrument 1 further comprises an output transducer L 1, which is configured to generate an output sound signal 6 from the reproduction signal A 1. The schematic representation of the hearing instrument 1 in FIG. 1 shows, as already mentioned, a so-called Behind-The-Ear hearing aid (BTE) with an earpiece 8, in which the output transducer L 1 is arranged, but the hearing instrument 1 is also conceivable as a design, in particular as an In-The-Ear hearing aid (ITE), an In-The-Canal hearing aid (ITC), a Completely-In-the-Canal hearing aid (CIC), a Receiver-In-the-Canal hearing aid (RIC), or in particular also as an earphone provided not exclusively or not primarily for the care of a hearing weakness.During operation of the hearing instrument 1, a classification of the external audio signal Ext with respect to various possible classes of audio content is now carried out in a manner still to be described. For this purpose, the signal processing device 4 has at least one artificial neural network (and preferably two artificial neural networks) which implement a first classifier or a second classifier. The artificial neural networks can run via an ASIC or another hard-wired circuit in the signal processing device 4, or else as a program element on a signal processor of the signal processing device 4 configured for this purpose.FIG. 2 schematically shows a block diagram of a signal flow in the hearing instrument 1 according to FIG. 1 for the aforementioned classification of the external audio signal Ext with respect to various possible classes of audio content. In the present case, acoustic features Fa-n are first ascertained at regular intervals (i.e. with regular updating) on the basis of external audio signals Ext. The acoustic features Fa-n preferably comprise a center-of-gravity frequency of an overall signal and / or a noise background, a modulation depth at a given modulation frequency, a stationarity, a signal level, a noise level for at least a given frequency range, an autocorrelation value for at least a given time delay, a strike or accent detection, and / or an onset detection.The acoustic features Fa-n are now fed to a first classifier DNN 1 and a second classifier DNN 1 as respective input variables. The first classifier DNN 1 and also the second classifier DNN 2 here each have an internal structure, not shown in detail, which in particular comprises an input layer (input layer), one or preferably a plurality of hidden layers (hidden layers) and an output layer (output layer), wherein the hidden layers in turn have at least one so-called density layer or a fully connected layer as well as a so-called gated recurrent unit layer. In particular, the respective set of acoustic features Fa-n, which are supplied to the first classifier DNN 1 as input variables, can differ from the set of acoustic features, which are supplied to the second classifier DNN 2, in some features (so that, for example, the second classifier DNN 2 still receives some additional acoustic features Fa-n as input variables, which were not used by the first classifier DNN 1 and vice versa).The first classifier DNN 1 now uses the acoustic features Fa-n for a first plurality of first classes C 1 a- cof audio content to determine a first probability value w 1 a- cfor a presence of the associated audio content in the external audio signal Ext. The first classes C1a-c are given in the present case by the first classes "music" (C1a), "speech" (C1b) and "other" (C1c), but could also comprise further classes of audio contents without a major change to said first classes C1a-c.The first probability values w1a-c hereby each quantify the probability with which the associated audio content of the relevant first class C1a-c is contained in the external audio signal Ext, i.e. a first probability value w1x=1 (x=a... c) means that the audio content of the associated first class C1x is securely contained in the external audio signal Ext, while a first probability value w1x=0 means that the audio content of the associated first class C1x is securely not contained in the external audio signal Ext. The individual first probability values w1a-c therefore do not have to be normalized to 1 in their sum over all first classes C1a-c (they are generally not). In particular, in the event of a sudden occurrence of an audio content of a specific class C 1 x(i.e. from w 1 x=0 to w 1 x≠0), the first probability values w 1 y(y≠x) of the other first classes C 1 ydo not necessarily have to change.A post-processing 9 can also be applied to the first probability values w1a-d determined as described, which can in particular comprise a temporal smoothing. In particular, at least one time constant of the temporal smoothing can depend on the respective first probability value w 1 a- dself, in that e.g. a shorter time constant for an "attack" and a longer time constant for a "release" are selected if the first probability value w 1 a- dis longer above a predefined limit value (for example. 0.5 or 0.6).Preferably, the first probability values w1a for "music" (C1a) and w1b for "speech" (C1b) are now used for generating the reproduction signal A1 by controlling a signal processing 7 of the external audio signal Ext to form the reproduction signal A1 as a function of said first probability values w1a, w1b. This can be effected in particular by a parameter of the signal processing 7, such as a gain factor for a frequency range important for formant speech, running at least for an interval of the respective first probability values w1a, w1b (for example an interval with a lower limit of 0.3 to 0.5 and an upper limit of 0.7 to 1) in monotone dependence of said first probability values w1a, w21. This includes, in particular, that in such an interval, with an increasing first probability value w2a for "speech" (C1a) (with otherwise constant first probability values w1a, w1c of the other first classes C1a, C1c), the external audio signal Ext enters the playback signal A1 with a higher signal gain in the relevant frequency range. In particular, however, the signal processing 7 can also be set on the basis of a further classification of the external audio signal Ext.For this purpose, in the exemplary embodiment according to FIG. 2, a check is now carried out as to whether the first probability value w1a of the first class "music" (C1a) exceeds a lower limit value th1 (y / n). If this is the case (y), said further classification of the external audio signal Ext is carried out by means of the second classifier DNN 2. For this purpose, the acoustic features Fa-n are fed to the second classifier DNN 2 as respective input variables.Based on these input variables, a second probability value w2a-d for a presence of the associated audio content in the external audio signal Ext now respectively determines for a second plurality of second classes C2a-d of audio content. The second classes C2a-d are given in the present case by the second classes "classic music / orchester" (C2a), "pop / rock" (C2b), "electro / techno" (C2c) and "other music" (C2d), but could also comprise further classes of music without a major change to said second classes C2a-d (with the exception of the second class "other music" / C2d).As in the case of the first probability values w1a-c, post-processing 9 can also be applied to the second probability values w2a-d determined as described. The second and second probability values w2a-d can now likewise be used for setting parameters of the signal processing unit 7 for generating the reproduction signal A1 from the external audio signal Ext. In this case, the corresponding explanations regarding the setting of the signal processing 7 apply in particular on the basis of the first probability values w1a-c. The setting on the basis of the second probability values w2a-d can be effected in particular by a more detailed spectral shaping which takes into account in particular a typical frequency spectrum of the respective audio content of the relevant second class C2a-d.FIG. 3 shows an alternative embodiment of the hearing instrument 1 to that of FIG. 1. The hearing instrument 1 is in this case provided as a binaural hearing system 10 with a first local hearing device HG 1 and a second local hearing device HG 2, which are to be worn by a user on his left or right ear. Each of the two local hearing devices HG 1, HG 2 can be configured here in particular like the hearing system 1 according to FIG. 1, i.e. in particular each comprise two input transducers, so that the data signal DS with the external audio signal Ext 2 is received in each local hearing device HG 1, HG 2. Each of the local hearing devices HG 1, HG 2 is also configured to carry out the classification of the external audio signal Ext according to corresponding audio contents, as illustrated with reference to FIG. 2, for which purpose, in the case of a transmission of the external audio signal according to separate stereo channels, each of the two local hearing devices HG 1, HG 2 preferably uses only acoustic features which have been determined with reference to the respective local channel of the external audio signal. In order to achieve uniform classification on both sides, however, the output of the respective local first classifier (not shown) is not used directly as a first probability value for the associated first class, but is initially transmitted (dashed lines) as a provisional first probability value w', w" via suitably configured communication means (not shown), such as the communication device 4 according to FIG. 1, to the respective other local hearing device HG 1, HG 2, so that both provisional first probability values w', w" generated completely locally in each of the two local hearing devices HG 1, HG 2 are present. On each side, the respective first probability value w1y(y=a... c) for each of the first classes C1y can then be determined according to the same rule, for example by averaging or as maximum or as minimum, from the provisional first probability values w', w".Although the invention has been illustrated and described in more detail by the preferred exemplary embodiment, the invention is not restricted by the disclosed examples and other variations can be derived therefrom by the person skilled in the art without departing from the scope of protection of the invention.List of reference characters1 Hearing instrument 2 Ambient sound 4 Signal processing unit 5 Communication device 6 Output sound signal 7 Signal processing 8 Earpiece 9 Post-processing 10 Binaural hearing system A1 Reproduction signal C1a-c First classes C2a-d Second classes DNN1 / 2 First / second classifier DS Data signal E1 / 2 First / second input signal Ext External audio signal Fa-n Acoustic features HG1 / 2 First / second local hearing device (of the hearing instrument) L1 Output transducer M1 / 2 First / second input transducer th1 Lower limit value w1a-c, w1y First probability values w2a-d Second probability values w', w'' Preliminary second probability value

Claims

Method for signal processing of an external audio signal (Ext) in a hearing instrument (1), - wherein a data signal (DS) which contains the external audio signal (Ext) is received by a communication device (5) of the hearing instrument (1), - wherein the external audio signal (Ext) is examined with respect to a first set of acoustic features (Fa-n), - wherein a first classifier (DNN1) for a first plurality of first classes (C1a-c) of audio contents, on the basis of the acoustic features (Fa-n) of the first set, in each case determines as input variables a first probability value (w1a-c) which indicates a probability for a presence of the respective first class (C1a-c) of audio contents in the external audio signal (Ext), and - wherein at least one parameter of a signal processing (7) of the hearing instrument (1), which generates a reproduction signal (A1) on the basis of the external audio signal (Ext), is set on the basis of at least one of the first probability values (w1a-c).Method according to Claim 1, wherein each first probability value (w1a-c) is determined in such a way that the associated probability for a presence of the respective first class (C1a-c) of audio contents in the external audio signal (Ext) is specified independently of the respective other first probability values (w1a-c).Method according to claim 1 or claim 2, wherein the acoustic features (Fa-n) of the first set are selected from the following features: centroid frequency of a total signal and / or a noise background and / or a band limited signal, variance of the centroid frequency, modulation depth at at least one given modulation frequency, modulation phase difference of at least two different frequency bands, stationarity, signal level, noise level for at least one given frequency range, autocorrelation value for at least one given time delay, beat detection, onset detection.Method according to one of the preceding claims, wherein the first classifier (DNN1) is implemented by a first artificial neural network in the hearing instrument (1).Method according to one of the preceding claims, wherein at least the classes "music" (C1a) and "speech" (C1b) are included as first classes (C1a-d).Method according to one of the preceding claims, - wherein, on the basis of at least some of the acoustic features (Fa-n) of the first set and on the basis of at least one of the first probability values (w1 a-c), a second classifier (DNN2) for a second plurality of second classes (C2a-d) of audio contents which are different from the audio contents of the first plurality of first classes (C1a-c) in each case determines a second probability value (w2a-d) which indicates a probability for a presence of the respective second class (C2a-d) of audio contents in the external audio signal (Ext), and - wherein the at least one parameter of the signal processing (7) of the hearing instrument (1) and / or a further parameter of the signal processing (7) of the hearing instrument (1) is set on the basis of at least one of the second probability values (w2a-d).Method according to Claim 6, - wherein the at least one of the first probability values (w1a-c) is used as an input variable for the second classifier (DNN2), and / or - wherein the second classifier (DNN2) is activated on the basis of the at least one of the first probability values (w1a-c).Method according to claim 6 or claim 7, wherein at least some of the following classes are included as second classes (C2a-d): "discussion", "single speaker / podcast", "listening book"; "classic music / orchester", "pop / rock", "electro / techno", "music with gesang".Method according to one of the preceding claims, wherein on the basis of the at least one parameter of the signal processing (7) of the hearing instrument (1), in particular source-based spectral shaping is controlled in order to form the reproduction signal (A1).Method according to one of the preceding claims, wherein temporal smoothing is applied to at least one first probability value (w1a-c), and wherein a time constant of said smoothing is set as a function of the same first probability value (w1a-c) at a preceding point in time.Method according to one of the preceding claims, wherein a binaural hearing system (10) with two local hearing aids (HG1, HG2) is used as hearing instrument (1), wherein a provisional first probability value (w', w") is determined in each of the two local hearing aids (HG1, HG2) for each of the first classes (C1a-c), and wherein the first probability value (w1a-c) is determined for at least some of the first classes (C1a-c) on the basis of the two associated provisional first probability values (w', w") of the two local hearing aids (HG1, HG2).Hearing instrument (1) comprising - a communication device (5) for receiving a data signal (DS) including an external audio signal (Ext), and - signal processing means (4) configured to implement at least one first classifier (DNN1), wherein the hearing instrument (1) is configured to perform the method according to any of the preceding claims.

Images