Patent Application
20180220243
The present invention relates to hearing aid systems. The present invention also relates to a method of operating a hearing aid system and a computer-readable storage medium having computer-executable instructions, which when executed carries out the method.
Generally a hearing aid system according to the invention is understood as meaning any system which provides an output signal that can be perceived as an acoustic signal by a user or contributes to providing such an output signal, and which has means which are used to compensate for an individual hearing loss of the user or contribute to compensating for the hearing loss of the user or contribute to compensating for the hearing loss. These systems may comprise hearing aids which can be worn on the body or on the head, in particular on or in the ear, and can be fully or partially implanted. However, some devices whose main aim is not to compensate for a hearing loss, may also be regarded as hearing aid systems, for example consumer electronic devices (televisions, hi-fi systems, mobile phones, MP3 players etc.) provided they have, however, measures for compensating for an individual hearing loss.
Within the present context a hearing aid may be understood as a small, battery-powered, microelectronic device designed to be worn behind or in the human ear by a hearing-impaired user. Prior to use, the hearing aid is adjusted by a hearing aid fitter according to a prescription. The prescription is based on a hearing test, resulting in a so-called audiogram, of the performance of the hearing-impaired user's unaided hearing. The prescription is developed to reach a setting where the hearing aid will alleviate a hearing loss by amplifying sound at frequencies in those parts of the audible frequency range where the user suffers a hearing deficit. A hearing aid comprises one or more microphones, a battery, a microelectronic circuit comprising a signal processor, and an acoustic output transducer. The signal processor is preferably a digital signal processor. The hearing aid is enclosed in a casing suitable for fitting behind or in a human ear. For this type of traditional hearing aids the mechanical design has developed into a number of general categories. As the name suggests, Behind-The-Ear (BTE) hearing aids are worn behind the ear. To be more precise, an electronics unit comprising a housing containing the major electronics parts thereof is worn behind the ear and an earpiece for emitting sound to the hearing aid user is worn in the ear, e.g. in the concha or the ear canal. In a traditional BTE hearing aid, a sound tube is used to convey sound from the output transducer, which in hearing aid terminology is normally referred to as the receiver, located in the housing of the electronics unit and to the ear canal. In some modern types of hearing aids a conducting member comprising electrical conductors conveys an electric signal from the housing and to a receiver placed in the earpiece in the ear. Such hearing aids are commonly referred to as Receiver-In-The-Ear (RITE) hearing aids. In a specific type of RITE hearing aids the receiver is placed inside the ear canal. This category is sometimes referred to as Receiver-In-Canal (RIC) hearing aids. In-The-Ear (ITE) hearing aids are designed for arrangement in the ear, normally in the funnel-shaped outer part of the ear canal. In a specific type of ITE hearing aids the hearing aid is placed substantially inside the ear canal. This category is sometimes referred to as Completely-In-Canal (CIC) hearing aids. This type of hearing aid requires an especially compact design in order to allow it to be arranged in the ear canal, while accommodating the components necessary for operation of the hearing aid.
Within the present context a hearing aid system may comprise a single hearing aid (a so called monaural hearing aid system) or comprise two hearing aids, one for each ear of the hearing aid user (a so called binaural hearing aid system). Furthermore the hearing aid system may comprise an external device, such as a smart phone having software applications adapted to interact with other devices of the hearing aid system, or the external device alone may function as a hearing aid system. Thus within the present context the term “hearing aid system device” may denote a traditional hearing aid or an external device.
It is well known within the art of hearing aid systems that the optimum setting of the hearing aid system parameters may depend critically on the given sound environment. It has therefore been suggested to provide the hearing aid system with a multitude of complete hearing aid system settings, often denoted hearing aid system programs, which the hearing aid system user can choose among, and it has even be suggested to configure the hearing aid system such that the appropriate hearing aid system program is selected automatically without the user having to interfere. One example of such a system can be found in U.S. Pat. No. 4,947,432.
This general concept of automatically selecting the appropriate hearing aid system program requires that any given sound environment can be identified as belonging to one of several predefined sound environment classes. Methods and systems for carrying out this sound classification are well known within the art. However, these methods and systems may be quite complex and require significant processing resources, which especially for hearing aid systems may be a problem. On the other hand it may be an even worse problem if the sound classification method or system is not precise and reliable and therefore prone to misclassifications, which may result in deteriorated sound quality and speech intelligibility or degenerated comfort for the hearing aid system user.
It is therefore a feature of the present invention to provide a method of operating a hearing aid system that provides precise and robust sound classification using a minimum of processing resources.
It is another feature of the present invention to provide a hearing aid system adapted to provide precise and robust sound classification using a minimum of processing resources.
The invention, in a first aspect, provides a method of operating a hearing aid system comprising the steps of: providing an electrical input signal representing an acoustical signal from an input transducer of the hearing aid system; providing a feature vector comprising vector elements that represent features extracted from the electrical input signal; providing a first multitude of sound environment base classes, wherein none of the sound environment base classes are defined by the presence of speech; processing a second multitude of feature vectors in order to determine the probability that a given sound environment base class, from said first multitude of sound environment base classes, is present in an ambient sound environment; selecting a current sound environment base class by determining the sound environment base class that provides the highest probability of being present in the ambient sound environment; determining a final sound environment class based on said selected current sound environment base class and a detection of whether speech is present in the ambient sound environment; setting at least one hearing aid system parameter in response to said determined final sound environment class; and processing the electrical input signal in accordance with said setting of said at least one hearing aid system parameter, hereby providing an output signal adapted for driving an output transducer of the hearing aid system.
The invention, in a second aspect, provides a non-transient computer-readable storage medium having computer-executable instructions, which when executed carry put the above method.
The invention, in a third aspect, provides a hearing aid system comprising a hearing aid processor adapted for processing an input signal in order to relieve a hearing deficit of an individual user, and a sound environment classifier (104), wherein the sound environment classifier further comprises a feature extractor, a base class classifier and a final class classifier, wherein the hearing aid processor or the sound environment classifier comprises a speech detector that is configured to provide information to the final class classifier whether speech is present or not in the sound environment.
Further advantageous features appear from the appended claims.
Still other features of the present invention will become apparent to those skilled in the art from the following description wherein the invention will be explained in greater detail.
By way of example, there is shown and described a preferred embodiment of this invention. As will be realized, the invention is capable of other embodiments, and its several details are capable of modification in various, obvious aspects, all without departing from the invention. Accordingly, the drawings and descriptions will be regarded as illustrative in nature and not as restrictive. In the drawings:
Reference is first made to
However, the hearing aid processor 103 also provides various features to the classifier 104 via the classifier input signal 114. The sound environment classification may therefore be carried out based on the input frequency band signals 111, the classifier input signal 114 and the broadband input signal 110.
Reference is now made to
The feature extractor 201 provides as output a multitude of extracted features that may either be derived from the broadband input signal 110, from the input frequency band signals 111 or from the hearing aid processor 103 via the classifier input signal 114.
According to the first embodiment of the invention the broadband input signal 110 is passed through the band-pass filter bank 102, whereby the input signal 110 is transformed into fifteen frequency bands 111 with center frequencies that are non-linearly spaced by setting the center frequency spacing to a fraction of an octave, wherein the fraction may be in the range between 0.1 and 0.5 or in the range between 0.25 and 0.35. One advantage of having this particular frequency band distribution is that it allows features that reflect important characteristics of the human auditory system to be extracted in a relatively simple and therefore processing efficient manner.
However in variations of the first embodiment the band-pass filter bank may provide more or fewer frequency bands and the frequency band center frequencies need not be non-linearly spaced, and in case the frequency band center frequencies are non-linearly spaced they need not be spaced by a fraction of an octave.
According to the first embodiment of the invention the extracted features from the feature extractor 201 comprises a variant of Mel Frequency Cepstral Coefficients, a variant of Modulation Cepstrum coefficients, a measure of the amplitude modulation, a measure of envelope modulation and a measure of tonality.
The variant of the Mel Frequency Cepstral Coefficients Xk, according to the present embodiment, is given as a scalar product of a first and a second vector, wherein the first vector comprises N elements xn, each holding an estimate of the absolute signal level, given in Decibel, of the signal output from a frequency band n provided by the filter bank 102 and wherein the second vector comprises N pre-determined values hn,k given by the formula:
wherein the index n represents a specific one of the input frequency bands 111, and wherein the scalar product is determined as a function of a selected specific value of k, such that the value of the k'th coefficient Xk is given by the Direct Cosine Transform (DCT):
This DCT is commonly known as DCT-II and in variations of the present embodiment other versions of a DCT may be applied.
These variants of the Mel Frequency Cepstral Coefficients are advantageous over the original Mel Frequency Cepstral Coefficients (MFCCs) with respect to the required processing resources in a hearing aid system.
Although original MFCCs may be found in slightly varying versions, all variants share some basic characteristics including the steps of:
Considering the differences between the variant of the MFCCs, according to the present embodiment, and the original MFCCs it follows that the steps 1)-3) described above may be omitted and instead replaced by the steps of applying the estimate of the absolute signal levels, given in Decibel, of the signal output from the frequency bands, which are determined anyway for other purposes by the hearing aid processor 103 and which therefore may be achieved directly from the hearing aid processor 103 using only a minimum of processing resources as opposed to having to carry out a Fourier transform, mapping the resulting spectrum onto the Mel scale and taking the logarithm of the power levels at each of the Mel frequencies.
In obvious variations of the first embodiment the estimate of the absolute signal level need not be given in Decibel. As one alternative other logarithmic forms may be used.
According to the first embodiment only the 2nd to 7th cepstral coefficients are extracted by the feature extractor 201. However, in variations of the first embodiment more or fewer cepstral coefficients may be extracted and in further variations all frequency bands need not be used for determining the cepstral coefficients.
According to the first embodiment the estimate of the absolute signal level xn used for determining the variant of the MFCCs is determined in accordance with the formula:
x
n(s)=xn(s−1)(1−α)+|yn(s)|α
wherein the index n represents a specific one of the input frequency bands 111, wherein s represents a discrete time step determined by a sample rate, wherein yn(s) represents samples of the absolute signal level, wherein α is a constant in the range between 0.01 and 0.0001 or between 0.005 and 0.0005, and wherein the sample rate is 32 kHz or in the range between 30 and 35 kHz. Obviously, the selected values of the sample rate and the constant α depend on each other in order to provide the estimate of the absolute signal level with the desired characteristics. In variations α may depend on the specific frequency band, since the signal variations and hereby the requirements to the absolute signal level estimate depends on the frequency range.
However, in variations other estimates of the absolute signal level may be used, e.g. the 90% percentile or a percentile signal in the range between 80% and 98%.
The variant of the modulation cepstrum coefficients is, as is the case for the cepstral coefficients, determined based on the input frequency bands 111 provided by the band-pass filter bank 102, and the final step of determining the modulation cepstrum coefficients is carried out by a calculating a simple scalar vector. In the following this variant of the modulation cepstrum coefficients may simply be denoted: modulation cepstrum coefficients. This variant of the modulation cepstrum coefficients is therefore advantageous for the same reasons as the cepstral coefficients according to the present embodiment.
More specifically the modulation cepstrum coefficients, according to the first embodiment of the invention, is determined by:
In variations of the first embodiment the feature representing the modulation cepstrum coefficients may be determined using other frequency ranges and/or more or less summed signals.
The feature representing the amplitude modulation may be determined in a variety of alternative ways all of which will be well known by a person skilled in the art and the same is true for the feature representing envelope modulation.
The feature extractor 201 also provides a feature representing tonality that may be described as a measure of the amount of non-modulated pure tones in the input signal. According to the embodiment of
However, a feature representing tonality may be determined in a variety of alternative ways all of which will be well known by a person skilled in the art.
It is a specific advantage of the present classifier 104 that a significant part of the features used to classify the sound environment are at least partly based on features that are calculated or determined for other purposes in the hearing aid system, whereby the amount of additional processing resources required by the classifier can be kept small.
According to the first embodiment of the invention a total of twelve features are provided from the feature extractor 201 and to the base class classifier 204 in the form of a feature vector with twelve individual elements each representing one of said twelve features. According to variations of the first embodiment of the invention fewer or more features may be included in the feature vector.
The base class classifier 204 comprises a class library, that may also be denoted a codebook. The codebook consists of a multitude of pre-determined feature vectors, wherein each of the pre-determined feature vectors are represented by a symbol. Additionally the base class classifier comprises pre-determined probabilities that a given symbol belongs to a given sound environment base class.
The pre-determined feature vectors and pre-determined probabilities that a given symbol belongs to a given sound environment base class are derived from a large number of real life recordings (i.e. training data) spanning the sound environment base classes. According to the present embodiment the base class classifier 204 is configured to have four sound environment base classes: urban noise, transportation noise, party noise and music, wherefrom it follows that none of the sound environment base classes are defined by the presence of speech.
Whenever a current feature vector is provided to the base class classifier 204, then the current feature vector is compared to each of the pre-determined feature vectors by using a minimum distance calculation to estimate the similarity between each of the pre-determined feature vectors and the current feature vector, whereby a symbol is assigned to each sample of the current feature vector, by determining the pre-determined feature vector that has the shortest distance to the current feature vector.
According to the present embodiment the codebook comprises 20 pre-determined feature vectors and accordingly there are 20 symbols.
According to the present embodiment the L1 norm also known as the city block distance is used to estimate the similarity between each of the pre-determined feature vectors and the current feature vector due to its relaxed requirements to processing power relative to other methods for minimum distance calculation such as the Euclidian distance also known as the L2 norm.
According to a variation of the present embodiment the training data are analyzed and the sample variance for each of the individual elements in the feature vector determined. Based on this sample variance the individual elements of a current feature vector are weighted such that the expected sample variance for each of the individual elements is below a predetermined threshold or within a certain range such as between 0.1 and 2.0 or between 0.5 and 1.5. However, since a weighting of data is involved the numerical value of the predetermined threshold can basically be anything. Obviously, the pre-determined feature vectors are weighted accordingly.
Hereby, it is avoided that a single element of the feature vector has a too high impact on the resulting distance to a pre-determined feature vector and furthermore the dynamic range required for the feature vector may be reduced, whereby the memory and processing requirements to the hearing aid system may likewise be reduced.
According to another variation of the present embodiment the training data are analyzed and the sample mean for each of the individual elements in the feature vector determined. Based on this sample mean the individual elements of a current feature vector are normalized, by subtracting the sample mean as a bias. In variations another bias may be subtracted, such that the expected sample mean for each of the individual elements is below a predetermined threshold of 0.1 or 0.5. However, since a weighting of data is involved the numerical value of the predetermined threshold may basically be anything. Obviously, the pre-determined feature vectors are normalized accordingly. Hereby, the dynamic range required for the feature vector may be reduced, whereby the memory and processing requirements to the hearing aid system may likewise be reduced.
It is a further advantage of the disclosed variations directed at weighting and normalizing the feature vector elements that the subsequent processing of the feature vector is simplified.
The 32 most recent identified symbols is stored in a circular buffer and by combining the stored identified symbols with the corresponding pre-determined probabilities that a given symbol belongs to a given sound environment base class, then a running probability estimate that a given sound environment base class is present in the ambient sound environment can be derived. The base class with the highest running probability estimate is selected as the current sound environment base class and provided to the final class classifier 205. According to the present embodiment the running probability estimate is derived by adding the 32 pre-determined probabilities corresponding to the 32 most recently identified symbols, wherein the pre-determined probabilities are calculated by taking a logarithm to the initially determined probabilities, which makes it possible to save processing resources because the pre-determined probabilities may be added instead of multiplied in order to provide the running probability estimate.
In variations fewer or more symbols may be stored, e.g. in the range between 15 and 50 or in the range between 30 and 35. By storing 32 symbols representing a time window of one second or in the range between a half and five seconds then an optimum compromise between complexity and classification precision is achieved.
According to another variation of the first embodiment of the invention an initial multitude of base classes and the corresponding running probability estimates are mapped onto a second smaller multitude of base classes. This allows a more flexible and precise sound environment classification because sound environments such as transportation noise may exhibit characteristics that are highly variable, e.g. dependent on whether a car window is open or closed. According to more specific variations the initial multitude of sound environment base classes comprises in the range between seven and fifteen base classes and the second smaller multitude comprises in the range between four and six sound environment base classes.
According to still other variations of the first embodiment of the invention the current base class that is provided to the final class classifier 205 is determined after low-pass filtering of the running probability estimates for each of the sound environment base classes. In variations other averaging techniques may be applied in order to further smooth the running probability estimates, despite that the implementation according to the first embodiment provides a smoothed output by summing the 32 pre-determined probabilities.
In addition to the current base class the final class classifier 205 also receives input from a speech detector 202 and a loudness estimator 203 and based on these three inputs the final sound environment classification is carried out.
The loudness estimator 203 provides an estimate that is either high or low to the final class classifier 205. The estimation includes: a weighting of the estimated absolute signal levels of the frequency band signals 111 in order to mimic the equal loudness contours of the auditory system for a normal hearing person, a summation of the weighted frequency band signal levels and a comparison of the summed levels with a predetermined threshold in order to estimate whether the loudness estimate is high or low. According to an advantageous variation the predetermined threshold is split into two predetermined thresholds in order to introduce hysteresis in the loudness estimation.
According to yet another variation the loudness estimation is determined by weighting the 10% percentile of the frequency band signals with the band importance function of a Speech Intelligibility Index (see e.g. the ANSI S3.5-1969 standard (revised 1997)) and selecting the largest weighted 10% percentile of the frequency band signals as the loudness level, that is subsequently compared with pre-determined thresholds in order to estimate the loudness as either high or low. It is a specific advantage of this variation that the largest level of the weighted 10% percentiles of the frequency bands is also used by the hearing aid system in order to determine an appropriate output level for sound messages generated internally by the hearing aid system.
It is a specific advantage of the present classifier 104 that the loudness estimation is carried out separately because this has made it possible to only apply features for the feature vector that are independent on the sound pressure level, whereby a more precise sound classification can be obtained.
The speech detector 202 provides an estimate of whether speech is present or not for the final class classifier 205. The speech detector may be implemented as disclosed in WO-A1-2012076045, especially with respect to
It is a specific advantage of the present classifier 104 that the speech detection is carried out separately because this allows the use of advanced methods of speech detection that operate independently of the remaining sound classification features, such as the feature extractor 201 and the base class classifier 204 according to the present embodiment. Hereby a more robust and precise sound classification can be obtained, because the sound environments representing the base classes are more distinctly different. Additionally the sound classification may require fewer processing resources because the feature vectors can be selected without having to include features directed at detecting speech. Yet another advantage according to the present embodiment is that the separate speech detection is carried out anyway by the hearing aid system and therefore requires basically no extra resources when being used by the classifier 104.
For reasons of clarity the speech detector 202 is illustrated in
According to the first embodiment of the present invention, the final class classifier 205 maps the current base class onto one of the final sound environment classes based on the additional input from the speech detector 202 and the loudness estimator 203, wherein the final sound environment classes represent the sound environments: quiet, urban noise, transportation noise, party noise, music, quiet speech, urban noise and speech, transportation noise and speech, and party noise and speech.
The mapping is carried out by first considering the loudness estimate, and in case it is low, the final sound environment class is quiet or quiet speech dependent on the input from the speech detector. If the loudness estimate is high then the final sound environment is selected as the current base class with or without speech again dependent on the input from the speech detector.
According to a variation of the first embodiment of the present invention, the input from the loudness estimator 203 and to the final class classifier 205 may be omitted and instead the loudness (i.e. the weighted sound pressure level) is included in the current feature vector, and in this case the sound environment base class will comprise the quiet sound environment.
According to yet another variation of the first embodiment of the present invention the final class classifier 205 additionally receives input from a wind noise detection block. If the wind noise detection block signals that the level of the wind noise exceeds a first predetermined threshold then the final sound environment class is frozen until wind noise again is below a second predetermined threshold. This prevents the classifier 104 from seeking to classify a sound environment that the classifier 104 is not trained to classify, and which sound environment is better handled by other processing blocks in the hearing aid system.
A first embodiment has been disclosed above along with a plurality of variations whereby multiple embodiments may be formed by including one or more of the disclosed variations in the first embodiment.
Reference is now made to
The method comprises:
The method embodiment of the invention may be varied by including one or more of the variations disclosed above with reference to the hearing aid system embodiment of the invention.
This application is a continuation of Application No. PCT/EP2015/072919 filed Oct. 5, 2015, the disclosure of which is incorporated by reference herein.
| Number | Date | Country | |
|---|---|---|---|
| Parent | PCT/EP2015/072919 | Oct 2015 | US |
| Child | 15938508 | US |
