EP1247425B1 - Method for operating a hearing-aid and a hearing aid - Google Patents

Abstract

The invention relates to a method for operating a hearing aid (1). The method is characterised in that characteristic features are extracted from an acoustic signal which has been recorded using at least one microphone (2a, 2b) and that said characteristic features are processed in an identification phase, using hidden Markov models, in particular for determining a current acoustic environmental scene or noises and/or for recognition of a speaker and words. The invention also relates to a hearing-aid.

Description

The present invention relates to a method for operating a hearing aid and a hearing aid.

Today, modern hearing aids can be adapted to different acoustic environment situations with the aid of various hearing programs-typically two to a maximum of three programs. Thus, the hearing aid should offer the user an optimal benefit in every situation.

The choice of hearing program can be made either via the remote control or via a switch on the hearing aid itself. However, switching between different programs is annoying or difficult, if not impossible, for many users. Which program offers the optimum comfort and the best speech intelligibility at which point in time is not always easy to determine even for experienced hearing aid wearers. An automatic recognition of the acoustic environment situation and an associated automatic switching of the hearing program in the hearing aid is therefore desirable.

Various methods for the automatic classification of acoustic environmental situations are currently known. In all of these methods, various features are extracted from the input signal that may originate from one or more microphones in the hearing aid. Based on these characteristics, a pattern recognizer, using an algorithm, makes a decision about the affiliation of the analyzed input signal to a specific acoustic environment situation. The various known methods differ on the one hand by the different features which in the description of the acoustic environment situation are used (signal analysis), and on the other hand by the pattern recognizer used which classifies the characteristics (signal identification).

For the feature extraction in audio signals was in the article of JM Kates entitled "Classification of Background Noises for Hearing-Aid Applications" (1995, Journal of the Acoustic Society of America 97 (1), pages 461-469 ) proposed to perform an analysis of the temporal level fluctuations and the spectrum. Furthermore, in the European patent specification number EP-B1-0 732 036 an analysis of the amplitude histogram to achieve the same goal proposed. Finally, the feature extraction was also investigated and applied by analyzing different modulation frequencies. In this regard, the two essays by Ostendorf et. al. entitled "Empirical Classification of Various Acoustic Signals and Speech by Modulation Frequency Analysis" (1997, DAGA 97, pages 608 to 609 ) and " Classification of acoustic signals based on the analysis of modulation spectra for use in digital hearing aids "(1998, DAGA 98, pages 402-403 ). A similar approach is also in an essay by Edwards et. al. entitled "Signal processing algorithms for a new software-based, digital hearing device" (1998, The Hearing Journal 51, pages 44 to 52 ) disclosed. Other possible features are the level itself or the zero crossing rate such as in HL Hirsch, "Statistical Signal Characterization" (Artech House 1992 ). The features used so far for audio signal analysis are therefore purely technical-based.

The known methods for noise classification, consisting of feature extraction and pattern recognition, have the disadvantages that, although a clear and robust identification of speech signals is basically possible, several different acoustic environment situations can not or only insufficiently classified. Thus, although it is possible with the known methods to be able to distinguish pure speech signals from "non-speech" - ie all other acoustic environment situations. However, this is not sufficient so that an optimal hearing program to be used for an instantaneous acoustic environment situation can be selected. As a consequence, either the number of possible hearing programs is limited to the two automatically recognizable acoustic environment situations or the hearing device wearer must recognize the uncovered acoustic environment situations himself and activate the associated hearing program by hand.

In principle, pattern identification methods known for noise classification can be used. Thus, in particular so-called distance estimators, Bayes classifiers, fuzzy logic systems or neural networks are suitable as pattern recognizers. Further information on the first two methods can be found in the publication " Pattern Classification and Scene Analysis "by Richard O. Duda and Peter E. Hart (John Wiley & Sons, 1973 ). Regarding neural networks, the standard work of Christopher M. Bishop entitled "Neural Networks for Pattern Recognition" (1995 Oxford University Press ). Further reference is made to the following publications: Ostendorf et. al., "Classification of Acoustic Signals Based on the Analysis of Modulation Spectra for Use in Digital Hearing Aids" (Journal of Audiology, 1998, pages 148 to 150 ); F. Feldbusch, "Noise Detection Using Neural Networks" (1998, Journal of Audiology, pages 30 to 36 ); European patent application with the publication number EP-A1-0 814 636 ; PCT application with the publication number WO 01 76321 A and U.S. Patent Publication Number U.S. 5,604,812 , However, all the mentioned pattern recognition methods have the disadvantage that they merely model static properties of the noise classes of interest.

Furthermore, it is off EP 0 881 625 a speech recognition system is known in which spoken words are translated into a written text. In this case, the spoken words (utterances) are recorded under a variety of reproduction conditions and filed as patterns, which are used as the basis for recognition in speech recognition using Hidden Markov models. The reproduction conditions are limited to the reproduction of the same word by different speakers with different ages, dialect, gender, pronunciation, etc.

The present invention is therefore based on the object of initially providing a method for operating a hearing aid, which is much more robust and accurate compared to the known methods.

This object is achieved by the measures specified in claim 1. Advantageous embodiments of the invention and a hearing aid are specified in further claims.

The invention is based on an extraction of signal features and a subsequent separation of different noise sources as well as an identification of different sounds, Hidden Markov models being used in the identification phase to detect a current ambient situation or noises and / or a speaker or his words. This is the first dynamic properties of the classes of interest takes into account, whereby a significant improvement in the accuracy of the inventive method in all applications, ie in the detection of the current environmental situation or noise as well as in the detection of a speaker and single words has been achieved.

In a further embodiment of the method according to the invention, auditory-based features are taken into account instead of or in addition to technically-based features in the extraction phase. These auditory-based features are preferably using methods of Auditory Scene Analysis (ASA).

In yet another embodiment of the method according to the invention, in the extraction phase a grouping of the features takes place context-independent or context-dependent with the aid of the gestalt principles.

The invention will be explained in more detail with reference to a drawing, for example. The single figure shows a block diagram of a hearing aid in which the inventive method is realized.

In the single figure, 1 denotes a hearing aid, whereby in the following the term "hearing aid" means both so-called hearing aids, which are used to correct a damaged hearing of a person, as well as all other acoustic communication systems, such as radios, are to be understood ,

The hearing aid 1 consists in a known manner initially of two electro-acoustic transducers 2a, 2b and 6, namely one or more microphones 2a, 2b and a speaker 6 - also referred to as a handset. An actual main component of a hearing aid 1 is a transmission unit designated 4 in which the - in the case of a hearing aid - adapted to the user of the hearing aid 1 signal modifications are made. However, the operations carried out in the transmission unit 4 are not only dependent on the type of a predetermined target function of the hearing aid 1, but are chosen in particular depending on the current acoustic environment situation. For this reason, z. B. already offered hearing aids, in which the equipment wearer can make a manual switch between different hearing programs on certain acoustic environment situations are adjusted. Likewise, hearing aids are known in which the detection of the acoustic environment situation is performed automatically. In this regard, let us reiterate the European patents with the publication numbers EP-B1-0 732 036 and EP-A1-0 814 636 as well as to the US patent with the publication number U.S. 5,604,812 , and the brochure "Claro Autoselect" by Phonak hearing systems ( 28148 (GB) / 0300 . 1999 ).

In addition to the mentioned components - such as microphones 2a, 2b, transmission unit 4 and handset 6 - a signal analysis unit 7 and a signal identification unit 8 are provided in the hearing aid 1. If the hearing device 1 is an implementation using digital technology, then one or more analog / digital converters 3a, 3b and between the transmission unit 4 and the earphone 6 are a digital / analog converter between the microphones 2a, 2b and the transmission unit 4 5 provided. Although the realization in digital technology is the preferred embodiment of the present invention, it is basically also conceivable that all components are realized in analog technology. In this case, it goes without saying that the transducers 3a, 3b and 5 are eliminated.

The signal analysis unit 7 is supplied with the same input signal as the transmission unit 4. Finally, the signal identification unit 8, which is connected to the output of the signal analysis unit 7, is connected to the transmission unit 4 and to a control unit 9.

Denoted by 10 is a training unit, with the aid of which the determination of required for the classification in the signal identification unit 8 parameters in an "off-line" operation is made.

The settings of the transmission unit 4 and the control unit 9 determined by the signal analysis unit 7 and the signal identification unit 8 can be overwritten by the user by means of a user input unit 11.

The following explains the method according to the invention:

A preferred embodiment of the inventive method is based on that in an extraction phase characteristic features are extracted from an acoustic signal, wherein instead of or in addition to technical-based features - such. For example, the previously mentioned zero-crossing rates, temporal level fluctuations, different modulation frequencies, or the level, the spectral centroid, the amplitude distribution, etc. - are also used auditory-based features. These auditory-based features are determined using Auditory Scene Analysis (ASA) and include in particular loudness, spectral form (timbre), harmonic structure (pitch), common settling and decay times (on / offsets), coherent amplitude modulations , coherent frequency modulations, coherent frequency transitions, binaural effects, etc. Explanations to the Auditory Scene Analysis are eg: B. in the works of A. Bregman, "Auditory Scene Analysis" (MIT Press, 1990 ) and WA Yost, "Fundamentals of Hearing - An Introduction" (Academic Press, 1977 ) to find. Information on the individual auditory-based characteristics can be found in, inter alia WAYost and S. Sheft, "Auditory Perception" (in "Human Psychophsics", edited by WA Yost, AN Popper and RR Fay, Springer 1993 ) WM Hartmann, "Pitch, periodicity, and auditory organization "(Journal of the Acoustic Society of America, 100 (6), pp. 3491-3502, 1996 ), such as DK Mellinger and BM Mont-Reynaud, "Scene Analysis" (in "Auditory Computation" edited by HL Hawkins, TA McMullen, AN Popper and RR Fay, Springer 1996 ).

As an example of the use of auditory-based features in signal analysis, the characterization of the tonality of the acoustic signal by analyzing the harmonic structure is given here, which is particularly suitable for the identification of tonal signals, such as speech and music.

In a further embodiment of the method according to the invention, it is provided to further group the features in the signal analysis unit 7 by means of Gestalt principles. In doing so, the principles of Gestalt theory, which examines qualitative qualities such as continuity, closeness, similarity, common destiny, unity, good continuation, and others, are applied to the auditory-based and possibly technical features for the formation of auditory objects. The grouping can - as well as the feature extraction in the extraction phase - either context-independent, ie without the addition of additional knowledge, carried out (so-called "primitive" grouping), or context-dependent in the sense of human auditory perception using additional information or hypotheses about the signal content (so-called "schema-based" grouping). The context-dependent grouping is thus adapted to the respective acoustic situation. For detailed explanations of the principles of Gestalt theory and grouping by means of gestalt principles, let me say for the following

Publications referenced: Perception Psychology "by EB Goldstein (Spektrum Akademischer Verlag, 1997 ), " Neural Bases of Gestalt Perception "by AK Engel and W. Singer (Spektrum der Wissenschaft, 1998, pp. 66-73 ), such as " Auditory Scene Analysis "by A. Bregman (MIT Press, 1990 ).

The advantage of using these grouping methods is that the characteristics of the input signal can be further differentiated. In particular, thereby signal parts are identifiable, which come from different sound sources. This allows the extracted features to be assigned to individual noise sources, thus providing additional knowledge about the existing noise sources - and thus about the current environmental situation.

The second aspect of the method according to the invention described here relates to pattern recognition or signal identification, which is carried out in the identification phase. In the preferred embodiment of the method according to the invention, the method of the Hidden Markov Models (HMM) is used in the signal identification unit 8 for automatic classification of the acoustic environment situation. Thus, temporal changes of the calculated characteristics can be used for classification. Consequently, dynamic and not only static properties of the environmental situations to be recognized resp. Noise classes are taken into account. Also possible is the combination of HMMs with other classifiers, eg. B. in a multi-stage detection method, to identify the acoustic environment.

According to the invention, the mentioned second aspect of the method, ie the use of hidden Markov models, is suitable. Excellent for determining a current acoustic environment or noise. Likewise, the recognition of a speaker as well as the detection of single words or phrases can be accomplished with extremely good results, even on their own, ie without the consideration of auditory-based characteristics in the extraction phase or even without the use of ASA. (Auditory Scene Analysis) methods, which is used in a further embodiment for determining characteristic features.

The output signal of the signal identification unit 8 thus contains information about the type of acoustic environment (acoustic environment situation). This information is applied to the transmission unit 4, in which the most suitable program for the detected environmental situation or the most suitable parameter set for the transmission is selected. At the same time, the information of the control unit 9 ascertained in the signal identification 8 is applied to further functions, where, depending on the situation, any function - e.g. an acoustic signal - can be triggered.

If Hidden Markov models are used in the identification phase, a complicated procedure for determining the parameters necessary for the classification becomes necessary. This parameter determination is therefore preferably carried out in an "off-line" procedure, for each class alone. The actual identification of different acoustic environment situations requires only small storage space and little computing capacity. Therefore, it is proposed to provide a training unit 9 which has sufficient computing power for parameter determination and which can be connected to the hearing aid 1 by suitable means for the purpose of data transfer. Such means may for example be a simple wire connection with be appropriate plugs.

With the method according to the invention, it is thus possible to select the most suitable from a multiplicity of different setting options and automatically retrievable actions without the user of the device having to take action himself. The comfort for the user is thus significantly improved, because it is selected immediately after the detection of a new acoustic environment situation, the right program or the corresponding function in the hearing aid 1 automatically.

Users of hearing aids often also wish to disable the above-described automatic recognition of the environmental situation and the associated automatic selection of the corresponding program. For this reason, an input unit 11 is provided, with which automatic responses or automatic program selection can be overwritten. Such input unit 11 may, for. B. a switch on the hearing aid 1 or a remote control, which is operated by the user.

Other possibilities such. As a voice-controlled user input are also conceivable.

Claims (14)

1. Method for operating a hearing device (1), the method consisting in

   - that, during an extraction phase, characteristic features are extracted from an acoustic signal recorded by at least one microphone (2a, 2b),

   - that, during an identification phase, the characteristic features are processed using the Hidden Markov Models for the determination of a momentary acoustic surrounding situation,

   characterized in

   - that, due to a determined momentary acoustic surrounding situation, a program or a transfer function, respectively, between at least one microphone (2a, 2b) and a receiver (6) is adjusted in the hearing device (1), such an adjustment being overwrite-able with the aid of a user input unit (11) by the hearing device user.
2. Method according to claim 1, characterized in that, for the determination of the characteristic features extracted during the extraction phase, ASA (Auditory Scene Analysis) methods are used.
3. Method according to claim 1 or 2, characterized in that at least one or several of the following auditory-based features are determined during the extraction of the features: loudness, spectral pattern, harmonic structure, common transient oscillation and decay events, coherent amplitude modulations, coherent frequency modulations, coherent frequency transitions and binaural effects.
4. Method according to one of the preceding claims, characterized in that additionally to auditory-based features also any other features are determined.
5. Method according to one of the preceding claims, characterized in that, for forming auditory objects, the auditory-based and, if need be, the other features are grouped along the principles of the Gestalt theory.
6. Method according to claim 5, characterized in that the extraction of the features and/or the grouping of the features is carried out either context-independently or context-dependently in terms of human auditory perception, in consideration of additional information or hypothesis on the signal content and thus adapted to the respective acoustic situation.
7. Method according to one of the preceding claims, characterized in that, during the identification phase, data is accessed that has been acquired during an "off-line" training phase.
8. Method according to one of the preceding claims, characterized in that the extraction phase and the identification phase take place continuously or at regular or irregular time intervals, respectively.
9. Method according to one of the preceding claims, characterized in that a certain function is triggered in the hearing device (1) in response to a detected momentary acoustic surrounding situation, a detected sound, a detected speaker or a detected word.
10. Hearing device (1) comprising a transfer unit (4), which is, on its input side, operatively connected to at least one microphone (2a, 2b), and which is, on its output side, operatively connected to a receiver (6),

    - the input signal of the transfer unit (4) being simultaneously fed to a signal analyzing unit (7) for the extraction of characteristic features,

    - the signal analyzing unit (7) being operatively connected to a signal identification unit (8), in which a momentary acoustic situation is determined using Hidden Markov Models,

    characterized in

    - that the signal identification unit (8) is operatively connected to the transfer unit (4) for adjusting a program or a transfer function, such an adjustment being over writable by the hearing device user with the aid of a user input unit (11).
11. Hearing device (1) according to claim 10, characterized in that an input unit (11) is provided which is operatively connected to the transfer unit (4).
12. Hearing device (1) according to claim 10 or 11, characterized in that a control unit (9) is provided, the signal identification unit (8) being operatively connected to the control unit (9).
13. Hearing device (1) according to claim 12, characterized in that the input unit (11) is operatively connected to the control unit (9).
14. Hearing device (1) according to one of the claims 10 to 13, characterized in that any means for transferring parameters from a training unit (10) to the signal identification unit (8) are provided.

Images