Hearing instrument and method of operating a hearing instrument

Abstract

A hearing instrument (2) and a method of operating the same is provided in order to alleviate sound artefacts due to sudden sound level changes.

Description

[0001] DESCRIPTION

[0002] Hearing instrument and method of operating a hearing instrument

[0003] The present invention relates to a hearing instrument. The invention also relates to a method of operating a hearing instrument.

[0004] In general, a hearing instrument is an electronic device being designed to support the hearing of person wearing it (which person is called the “user” or “wearer” of the hearing instrument). In particular, the invention relates to hearing instruments that are specifically configured to at least partially compensate a hearing impairment of a hearing-impaired user. Such hearing instruments are also called “hearing aids”. In addition to such hearing aids, there are hearing instruments that are designed to support the hearing of normal -hearing users (i.e. persons without a hearing impairment). Such hearing instruments, being sometimes referred to as “Personal Sound Amplification Products” (PSAP), may be provided, e.g., to enhance the hearing of the wearer in complex acoustic environments or to protect the hearing of the wearer from damage or overstress.

[0005] Hearing instruments, in particular hearing aids, are typically designed to be worn in or at an ear of the user, e.g. as a Behind-The-Ear (BTE) or In-The-Ear (ITE) device. With respect to its internal structure, a hearing instrument normally comprises an (acoustoelectrical) input transducer, a signal processor, and an output transducer. During operation of the hearing instrument, the input transducer captures a sound signal from an environment of the hearing instrument and converts it into an input audio signal (i.e. an electrical signal transporting a sound information). In the signal processor, the input audio signal is processed, in particular amplified dependent on frequency, e.g., to compensate the hearing-impairment of the user. The signal processor outputs the processed signal (also called output audio signal) to the output transducer. Most often, the output transducer is an electro-acoustic transducer (also called “receiver”) that converts the output audio signal into a processed airborne sound which is emitted into the ear canal of the user. Alternatively, the output transducer may be an electro-mechanical transducer that converts the output audio signal into a structure-borne sound (vibrations) that is transmitted, e.g., to the cranial bone of the user.

[0006] RestrictedFurthermore, besides classical hearing aids, there are implanted hearing aids such as cochlear implants, and hearing instruments the output transducers of which directly stimulate the auditory nerve of the user.

[0007] In order to provide frequency-dependent signal processing, the input audio signal output by the input transducer is normally fed to an analysis filter bank of the signal processor which splits the input audio signal into multiple frequency intervals, also referred to as frequency bands, wherein the signal in each of the frequency bands can be processed independently.

[0008] A crucial aspect in the design of a hearing instrument is the spectral resolution of the analysis filter bank, i.e. the number and spectral width of the frequency bands. On the one hand, a high resolution filter bank is desirable to achieve a high quality and flexibility of the signal processing. In particular, the quality of noise reduction and speech enhancement schemes normally benefits from a higher frequency resolution. However, on the other hand, a high frequency resolution generally increases processing time and, thus, leads to significant delay of the output audio signal as compared to the input audio signal. When output to the user, the delay of the output audio signal may have a detrimental effect on sound quality, in particular as the delay normally varies with the frequency of the frequency band signals. Moreover, the delay may negatively influence subsequent processing steps such as (dynamic) compression, beam forming, noise cancelling, noise suppression and feedback-cancelling. For example, for vowel onsets in a captured speech sound the analysis domain may lag up to ca. 15 ms after the original signal. It has been observed that for high compression ratios in the onset of vowels from silence the signal processing is not able to compress in time. The compression will thus effectively boost the onset leading to large sound levels and distortions of the processed sound in the form of a booming sound.

[0009] It is therefore an object of the present invention to provide solutions for signal processing in a hearing instrument with a high quality, in particular with a high spectral resolution of the signal analysis path but low delay and delay -induced distortions of the processed signal.

[0010] RestrictedAccording to a first aspect of the invention, the above object is met by a method as defined by claim 1 for operating a hearing instrument. The above-mentioned object is also met, according to a second aspect of invention, by a hearing instrument as defined by claim 9. Preferred embodiments of the invention are described in the dependent claims and the subsequent description.

[0011] In accordance with the method for operating a hearing instrument, a sound signal is captured from an environment of the hearing instrument. Said captured sound (in its original form as output by the at least one input transducer of the hearing instrument or after one or more optional preprocessing steps) is split into a number of first frequency band signals, using a first filter bank. Said first filter bank, subsequently referred to as the “analysis filter bank”, thus transforms the captured sound signal from time domain to frequency domain (by providing a plurality of frequency band signals).

[0012] In the frequency domain, the captured sound signal is processed by applying at least one processing step to the first frequency band signals or signals derived therefrom to (directly or indirectly) provide a processed sound signal. Preferably, said processing step is a delay-sensitive processing step, wherein “delay-sensitive” means that the processing step is sensitive to a temporal delay of the first frequency band signals or signals derived therefrom as compared to the captured sound signal in that such temporal delay results in a reduced quality of the processed sound signal. In preferred embodiments, the at least one delay-sensitive processing step is selected from one of (dynamic) compression, beam forming, noise cancelling, noise suppression, speech enhancement and feedback-cancelling.

[0013] The processed sound signal is then output to a user of the hearing instrument, e.g. by converting said processed sound signal or a signal derived therefrom into air-borne sound, using a receiver as the output transducer of the hearing instrument, and emitting said air-borne sound into the ear canal of the user.

[0014] In order to avoid or at least reduce detrimental effects of signal processing latency, the method includes an additional step of the splitting, by a second filter bank, the captured sound signal into a number of second frequency band signals, wherein

[0015] each of the second frequency band signals has a lower delay vis-a-vis the captured

[0016] Restrictedsound signal (I) or a signal derived therefrom (I’) compared to the delay vis-a-vis the captured sound signal (I) for at least one spectrally matching first frequency band signal (BA). Preferably, each of the second frequency band signals has a lower delay as compared to any spectrally matching first frequency band signal. Herein, the term “spectrally matching” means that a respective second frequency band has a same or similar center frequency (i.e. mean frequency) as a first frequency band with which it is compared. In particular, a first frequency band is called “spectrally matching” with a second frequency band, if the center frequency of said first frequency band is within the frequency range of said second frequency band.

[0017] Subsequently to splitting the captured sound signal into the first and second frequency band signals, the method comprises a comparison step in which each of all or at least a subset of the second frequency band signals are compared (directly or indirectly) with one spectrally matching first frequency band signal or with a signal derived from a plurality of spectrally matching first frequency band signals to receive a comparison result. Preferably, said comparison result is received by a subtraction of the respective signals to be compared or signals derived therefrom.

[0018] In further steps, the method comprises calculating a (frequency-dependent) correction gain based on the comparison result; and applying said correction gain to the first frequency band signals or signals derived therefrom (in particular by multiplication) to receive corrected first frequency band signals. Herein, said at least one (in particular delaysensitive) processing step is applied to the corrected first frequency band signals.

[0019] The idea of comparing the first frequency band signals with low-delay second frequency band signals and correcting the first frequency band signals by virtue of the comparison result was shown to be a very effective way to avoid or at least reduce the detrimental effects of the delay caused by the analysis filter bank to the subsequent signal processing. In particular, the method was proven to efficiently avoid or at least reduce distortions (booming sounds) at the vowel onset of processed speech sound. Therefore, without excluding other uses, effects and benefits, the second filter bank is also referred to as the “onset handling filter bank” (OH filter bank).

[0020] RestrictedIn preferred embodiments of the invention, the reduced delay of the second frequency band signals as compared to the at least one spectrally matching first frequency band signal is achieved by designing the second filter bank with a lower spectral resolution as compared with the first filter bank. In these embodiments, thus, the number of first frequency bands exceeds the number of second frequency bands. Preferably, the number of first frequency bands exceeds the number of second frequency bands by a factor of at least 4, more preferred a factor of at least 8, and in particular a factor of at least 10. In preferred embodiments, the first filter bank is realized as a warped filter bank. Such filter banks are known per se, e.g. from S. Wabnick, et al., “Frequency-warping in low delay audio coding”, ICASSP 2005, 2005 (IEEE), ISBN: 0-7803-8874-7.

[0021] In a further embodiment of the invention, in order to achieve low latency, the OH filter bank is designed such that the number of second frequency bands is below 10, in particular less than or equal to 6.

[0022] Preferably, the second frequency band signals or signals derived therefrom are temporally smoothed prior to the comparison step. Such smoothing is of particular benefit in embodiments of the invention in which the second frequency band signals or signals derived therefrom are down-sampled (i.e. reduced with respect to their temporal resolution) prior to the comparison step. In the latter case, the smoothing removes or at least reduces a modulation of the second frequency band signals or signals derived therefrom caused by the down-sampling.

[0023] In a particular beneficial embodiment of the invention, a Kalman filter is used for said smoothing (in the following this may also be denoted a Kalman smoother). For the Kalman smoother, the second frequency band signals or signals derived therefrom are assumed to be stochastic processes (i.e. signals with a temporally constant mean value and a gaussian random variation).

[0024] The Kalman smoother combines past and future observations in order to estimate system states more accurately. More specifically the Kalman Smoother is adapted to carry out the following steps:

[0025] - determining the Kalman Gain (K), as:

[0026] RestrictedK = p(i-1) / (P(i-1)+R(i-1))

[0027] P(i— 1)+P(i— 1)’

[0028] wherein P(i-l) is the previous prediction uncertainty and R(i-l) the previous measurement variance and wherein the Kalman gain is used to determine the weight given to a new measurement versus the predicted one.

[0029] - determining the Prediction Error (e) as:

[0030] e = in(i) — out(i — 1)

[0031] wherein the prediction error (e) is the difference between the actual input in(i) and the predicted output out(i — 1). Thus the prediction error (e) indicates how far off the prediction was from the actual measurement. By determining the prediction error (e), the Kalman smoother can adjust its predictions to reduce future errors, improving the accuracy of the state estimates.

[0032] - updating the prediction as:

[0033] out(i) = out(i — 1) + K * e;

[0034] wherein the updated prediction (state estimate) out(i) is determined by the previous prediction out(i-l) adjusted by the Kalman gain K and the prediction error e. The prediction error e is used to update the state estimate (i.e. the predicted output) (out(i)). If the error is large, it means the prediction was not accurate, and the state estimate needs significant adjustment. Conversely, if the error is small, the prediction was close to the actual measurement, and only a minor adjustment is needed Thus the prediction error provides a feedback mechanism that allows the Kalman smoother to continuously improve its predictions. By learning from past errors, the smoother can make more accurate future predictions. In other words the predicted output out(i-l) represents the Kalman smoother’s estimate of the system state at the previous time step. By comparing this prediction with the actual input in(i). the smoother can determine the prediction error and use it to refine its estimates. This process helps in reducing the overall error and improving the accuracy of the state estimates over time.

[0035] RestrictedThus the Kalman gain (K) determines how much the prediction error should influence the updated state estimate. It balances the trust between the prediction and the new measurement.

[0036] - updating the measurement variance as:

[0037] 7

[0038]

[0039] R(i) = (1 - a) * R(i - 1) + a * ((R_h)² + covar_r0)

[0040] wherein the updated measurement variance R(i) is determined by combining the previous measurement variance R(i-1 ) with the squared prediction error Rh2and an additional variance term cover rO, weighted by a smoothing factor a. According to an embodiment the prediction error Rhis filtered by an IIR filter, whereby the performance of the Kalman smoother can be further improved, requiring only a minor amount of additional processing.

[0041] - updating the prediction uncertainty by:

[0042] P(i) = (1 — K) * P(i — 1) + covar_q

[0043] wherein the prediction uncertainty P(i) is updated based on the previous prediction uncertainty P(i — 1) based on the Kalman gain K and the process noise covariance covar_q.

[0044] By leveraging both past and future data, the Kalman smoother provides more accurate state estimates compared to e.g. the Kalman filter, which only uses past and current data. The Kalman smoother therefore ensures that the state estimates are dynamically adjusted based on the latest available data, making the smoother adaptive to changes in the system. By incorporating the prediction error, the Kalman smoother can filter out noise and focus on the true signal, leading to more reliable state estimates.

[0045] In some embodiments of the invention, the first and second frequency band signals are not compared directly. Instead, in preferred embodiments, the comparison result is received by determining the power of the second frequency band signals and (for each of these) the power of a spectrally matching first frequency band signal or a plurality of spectrally matching first frequency band signals, and by comparing said power values

[0046] Restrictedwith each other (in particular by subtraction), wherein preferably said power values are obtained simply by squaring the values of the frequency band signals.. Herein, preferably, the power of the second frequency band signals is temporally smoothed prior to the comparison step, in particular using a Kalman smoother as already described above. In further embodiments of the invention, the power of the (smoothed) second frequency band signals is down-sampled prior to the comparison step.

[0047] In preferred embodiments of the invention, the correction gain is calculated according to formulas

[0048] G(t,f_A) = Σ m_OA(f_A,f_O) · G'(t,f_O)

[0049] Eq. 1

[0050] G'(t,f_O) = 1 +

[0051]

[0052] P'_A(t,f_O)

[0053] Eq. 2

[0054] ΔP(t,f_O) = P_O(t,f_O) - P'_A(t,f_O)

[0055] Eq. 3

[0056] and

[0057] P'_A(t,f_O) = Σ m_AO(f_O,f_A) · P_A(t,f_A)

[0058]

[0059] fA

[0060] Eq. 4

[0061] Herein

[0062] - G = G(t,fA) is the correction gain to be calculated,

[0063] - AP = AP(t,fo) is the comparison result,

[0064] - PA= PA(t,fA) is the power of the first frequency band signal at frequency fA,

[0065] Restricted- Po = Po(t,fo) is the (optionally smoothed and / or down-sampled) power of the sec- ond frequency band signal at frequency fo,

[0066] - m_AO(f_O,f_A) is an element of a first mapping function M_AO that maps the respective power P_A(t,f_A) of the first frequency band signals to the frequencies f_O of the second frequency band signals, and

[0067] - m_OA(f_A,f_O) is an element of a second mapping function M_OA that maps correction gains calculated for the frequencies f_O of the second frequency band signals to the frequencies f_A of the first frequency band signals, and is an optional correction function to be multiplied with the comparison result.

[0068] In Eq. 1 to 4, f_A and f_O denote the center frequencies of the first frequency bands and the second frequency bands, respectively. In a mathematical implementation, G(t,f_A), P_A(t,f_A) and P_O(t,f_O) can be represented by components of vectors (which vectors have a component for each frequency f_A or f_O, respectively). In this mathematical representation, the mapping functions MAO and MOA are represented by matrices having an element mAo(fo,fA) and moA(fA,fo) for each pair of frequencies \ and fo. Thus, M_AO and M_OA have a size of N_O × N_A and N_A × N_O, respectively, where NA is the number of first frequency bands and No is number of second frequency bands; if not all first or second frequency bands are used in the comparison, then NA and No are the number of first or second frequency bands used in the comparison. The variable t denotes time or, more specifically, a discrete time variable denoting a time frame. Thus G(t,fy), G’(t,fo), P\(t,f\), P’A(t,fo), Po(t,fo) and AP(t,fo) are time dependent quantities.

[0069] Preferably, the elements of the mapping functions MAO and MOA are chosen such that MAO is inverse to MOA (M_OA = M_AO⁻¹). Moreover, preferably, the elements of the mapping functions MAO and MOA are normalized such that the condition:

[0070] Σ_{f_A} m_OA(f_A,f_O) = Σ_{f_O} m_AO(f_O,f_A) = 1

[0071]

[0072] fA fo

[0073] Eq. 5

[0074] Restrictedis fulfilled.

[0075] The correction function C represents an optional correction. It may thus be set to 1 (i.e. C = 1). However, in preferred embodiments of the invention, the correction function C is defined such that small values of the comparison result AP(t,fo) are attenuated, in which cases the correction function has a value of zero (i.e. C = 0), whereas larger values of AP(t,fo) are passed unchanged, in which cases the correction function has a value of 1 (C = 1). Herein, preferably, the correction function C is implemented as step function, e.g. as

[0076] C(&p(t, fo)) = f° < MMo)

[0077]

[0078] 1 otherwise

[0079] Eq. 6

[0080] As an alternative to the hard step function according to Eq. 6, within the scope of the invention, a smooth step function such as a sigmoid function may be used as the correction function C.

[0081] Preferably, the step value APo(fo) of this function, i.e. the value of AP(t,fo) at which the correction function C changes from zero to one, is defined such that it is equal to or at least approximately corresponds to the covariance of the two filter banks power estimates P’A and Po (i.e. APo(fo) = COV P’A, PO)). This implementation makes sure, that the correction function assumes non-zero values only when a difference of P’A(t,fo) and Po(t,fo) is due to time alignment rather than noise

[0082] In accordance with the invention, the processed signal to be output to the user may be provided by recombining the corrected and processed first frequency band signals in a synthesis filter bank. However, preferably, the processed signal is derived by applying an adaptive (time-domain) filter to the captured signal in the time domain, wherein at least one filter coefficient of said time-domain filter is adapted based on the corrected and processed first frequency band signals. Application of such time-domain filters for signal processing in hearing instruments is known, per se, e.g. from US 11,310,607 B2 and US 2023 / 0026692 Al, the disclosure of which is incorporated in this application by reference. Preferably, the time-domain filter is realized as a Finite-Impulse-Response (FIR) filter.

[0083] RestrictedThe hearing instrument according to the second aspect of the invention comprises

[0084] - an input transducer arranged to capture a sound signal from an environment of the hearing instrument;

[0085] - an output transducer configured to output a processed sound signal to a user of the hearing instrument;

[0086] - a first filter bank configured to split said captured sound signal or a signal derived from it into a number of first frequency band signals; and

[0087] - a signal processor configured to apply at least one processing step to the first frequency band signals or signals derived therefrom to contribute to providing the processed sound signal.

[0088] Generally, the hearing system is configured to automatically perform the method according to the first aspect of the invention. To this end, the hearing instrument further comprises:

[0089] - a second filter bank configured to split the captured sound signal into a number of second frequency band signals, wherein each of the second frequency band signals has a lower delay vis-a-vis the captured sound signal as at least one (preferably any) spectrally matching first frequency band signal;

[0090] - an onset enhancement module configured to

[0091] - compare, in a comparison step, each of at least a subset of the second frequency band signals with one spectrally matching first frequency band signal or a signal derived from a plurality of spectrally matching first frequency band signals to receive a comparison result;

[0092] - calculate a (frequency-dependent) correction gain based on the comparison result; and

[0093] - applying said correction gain to the first frequency band signals to receive corrected first frequency band signals.

[0094] Herein, the signal processor is configured to apply the at least one (in particular delaysensitive) processing step to the corrected first frequency band signals.

[0095] RestrictedAs the hearing instrument according to the second aspect of the invention is configured to automatically perform the method according to the first aspect of the invention, each embodiment or variation of said method corresponds to an embodiment or variation of the hearing instrument. Hence, disclosure related to the method also applies, mutatis mutandis, to the hearing instrument, and vice-versa.

[0096] In particular, in preferred embodiments of the hearing instrument, the first filter bank (analysis filter bank) and the second filter bank (onset enhancement filter bank) are designed such that the number of first frequency bands exceeds the number of second frequency bands. Preferably, the number of first frequency bands exceeds the number of second frequency bands by a factor of at least 4, preferably a factor of at least 8, and in particular a factor of at least 10. In particular, the number of second frequency bands is below 10, in particular less than or equal to 6.

[0097] Preferably, the onset enhancement module is configured to (temporally) smooth the second frequency band signals or signals derived therefrom, in particular using a Kalman-Filter (which in the present context may also be denoted a Kalman smoother), prior to the comparison step.

[0098] In further preferred embodiments of the hearing instrument, the onset enhancement module is configured to compare (in particular subtract), in the comparison step, power values derived from the at least subset of second frequency band signals and the at least one spectrally matching first frequency band signal, respectively. Herein, preferably, the onset enhancement module is configured to (temporally) smooth and / or down-sample the respective power derived from the second frequency band signals prior to the comparison step.

[0099] In further preferred embodiments of the hearing instrument, the onset enhancement module is configured to calculate the correction gain according to Eqs. 1 to 4 as described above.

[0100] Preferably, the signal processor is designed as a digital electronic device. It may be a single unit or consist of a plurality of sub-processors. In accordance with the invention, the signal processor or at least one of said sub-processors may be a programmable device

[0101] Restricted(e.g. a micro-controller). In this case, the functionality mentioned above or part of said functionality is implemented as software (in particular firmware) which is stored in executable form in a storage of the signal processor. Also, the signal processor or at least one of said sub-processors may be a non-programmable device (e.g., an ASIC). In this case, the functionality mentioned above or part of said functionality is implemented as hardware circuitry.

[0102] Preferably, the onset enhancement module is arranged in the hearing instrument. In particular, the onset enhancement module may be designed as a hardware or a software component of the signal processor or as a separate electronic component.

[0103] Subsequently, embodiments of the present invention will be described in more detail with reference to the accompanying drawings in which

[0104] Fig. 1 shows a schematic representation of a hearing instrument comprising an input transducer arranged to capture a sound signal from an environment of the hearing instrument, a signal processor arranged to process the captured sound signal, and an output transducer arranged to emit the processed sound signal to a user; and

[0105] Fig. 2 shows a schematic representation of a part of the functional structure of the signal processor of the hearing instrument shown in Fig. 1.

[0106] In the figures, like reference numerals always indicate like parts, structures and elements unless indicated otherwise.

[0107] Fig. 1 shows a hearing instrument 2 that is configured to be worn in or at one of the ears of the user. Preferably, the hearing instrument 2 is a hearing aid, i.e. a hearing instrument being configured to support the hearing of a hearing-impaired user. As shown in Fig. 1, by way of example, the hearing instrument 2 may be designed as a Behind-The-Ear (BTE) hearing instrument. However, in alternative embodiments of the invention, the hearing instrument 2 may be realized in any of the further structural shapes mentioned above, e.g. as a receiver-in-canal device or an in-the-ear device. Optionally, the hearing instrument 2 may be part of a hearing system comprising at least one of a second hearing instrument (not shown) to be worn in or at the other ear of the user to provide binaural

[0108] Restrictedsupport to the user, an external programming device, a remote control, a charger and software application for controlling and / or programming the hearing instrument.

[0109] The hearing instrument 2 comprises, inside a housing 4, two microphones 6 as input transducers and a receiver 8 as output transducer. The hearing instrument 2 further comprises a battery 10 and a signal processor 12. Preferably, the signal processor 12 comprises both a programmable sub-unit (such as a microprocessor) and a non-programma-ble sub-unit (such as an ASIC).

[0110] The signal processor 12 is powered by the battery 10, i.e., the battery 10 provides an electric supply voltage U to the signal processor 12.

[0111] During normal operation of the hearing instrument 2, the microphones 6 capture an airborne sound from an environment of the hearing instrument 2. The microphones 6 convert the air-borne sound into an input audio signal I (also referred to as the “captured sound signal”), i.e., an electric signal containing information on the captured sound. The input audio signal I is fed to the signal processor 12. The signal processor 12 processes the input audio signal I, i.a., to provide a directed sound information (beam-forming), to perform noise reduction and dynamic compression, and to individually amplify different spectral portions of the input audio signal I based on audiogram data of the user to compensate for a user-specific hearing impairment. The signal processor 12 emits an output audio signal O (also referred to as the “processed sound signal”), i.e., an electric signal containing information on the processed sound to the receiver 8. The receiver 8 converts the output audio signal O into processed air-borne sound that is emitted into the ear canal of the user, via a sound channel 14 connecting the receiver 8 to a tip 16 of the housing 4 and a flexible sound tube (not shown) connecting the tip 16 to an ear piece inserted in the ear canal of the user.

[0112] As illustrated in Fig. 2, the signal processor 12 includes

[0113] - a (first) filter bank 20 that is also referred to as the “analysis filter bank”,

[0114] - a (second) filter bank 22 that is also referred to as the “onset enhancement filter bank” or “OH filter bank”,

[0115] - a first power determination module 24,

[0116] Restricted- a second power determination module 26,

[0117] - a smoother 28,

[0118] - a down-sampler 30,

[0119] - a first mapper 32,

[0120] - a second mapper 34,

[0121] - an adder 36

[0122] - a correction module 38,

[0123] - a multiplier 40,

[0124] - a compressor 42,

[0125] - a gain handling module 44,

[0126] - a filter engine 46,

[0127] - an interpolation engine 48, and

[0128] - an adaptive time-domain filter 50 that, preferably, is realized as a Finite Impulse Response (FIR) filter.

[0129] Preferably, the signal processor 12 further includes a first spatial analysis module 52 operating in time domain and including a time-domain beam former, and a second spatial analysis module 54 operating in frequency domain (i.e. based on frequency bands) and including a beam former.

[0130] Together, the OH filter bank 22, the power determination modules 24 and 26, the smoother 28, the down-sampler 30, the mapper 32 and 34, the adder 36, the correction module 38 and the multiplier 40 form an onset enhancement module 56 which, in the exemplary embodiment, is implemented as a functional part of the signal processor 12.

[0131] In a preferred implementation of the signal processor 12, the filter banks 20 and 22, the power determination modules 24 and 26, the smoother 28, the down-sampler 30, the first mapper 32, the second mapper 34, the adder 36, the correction module 38, the multiplier 40, the compressor 42, the gain handling module 44, the filter engine 46, the interpolation engine 48, the time-domain filter 50, and the spatial analysis modules 52 and 54 are realized as software modules. In alternative implementations, one or more of these components, in particular at least one of the filter banks 20 and 22 and / or the time domain filter 50, are be realized as hardware circuitry.

[0132] RestrictedDuring operation of the hearing instrument 2, in the time domain, the input audio signal I of the microphones 6 is fed to the spatial analysis module 52 that, inter alia, applies time-domain beamforming to the input audio signal I. The spatial analysis module 52 outputs a directed (time-domain) audio signal I’ to the time-domain filter 50. The timedomain filter 50 modifies the directed audio signal I’, thereby creating a processed audio signal P. As compared to the input audio signal I, the processed audio signal P is modified so to support the hearing of the user, in particular to at least partially compensate for a hearing impairment of the user. To this end, the time-domain filter 50 is adapted in dependency of a set of filter coefficients F that are created based on analysis of the input audio signal I in the frequency domain (e.g. by being based on frequency bands).

[0133] For said analysis, the input audio signal I of the microphones 6 is also fed to the analysis filter bank 20. The analysis filter bank 20 splits the input audio signal I of each microphone 6 in a number of (first) frequency band signals BA. The number of first frequency band signals BA is denoted NA. The center frequencies (i.e. mean frequencies) of the first frequency band signals BA are denoted f\.

[0134] The respective first frequency band signals BA of the two microphones 6 are fed to the second spatial analysis module 54 that applies beam forming to the first frequency band signals BA, thereby creating a corresponding number of NA directed first frequency band signals BA’.

[0135] The directed audio signal I’ is also fed to the OH filter bank 22 that splits said directed audio signal I’ in a number of (second) frequency band signals Bo. The number of second frequency band signals Bo is denoted No.

[0136] Preferably, OH filter bank 22 has a significantly lower spectral resolution but a significantly higher temporal resolution as compared to the analysis filter bank 20. By way of example, without loss of generality, it will be assumed in the following that the analysis filter bank 20 is designed to split the respective input audio signal I of the two microphones 6 in 65 first frequency band signals BA (i.e. NA = 65) with a temporal resolution of 2 kHz, whereas the OH filter bank 22 is designed to split the directed audio signal I’ into 6 second frequency band signals Bo (i.e. No = 6) with a temporal resolution of 32 kHz. Preferably, at least the analysis filter bank 20 is implemented as a warped filter bank

[0137] Restrictedand in that case also the OH filter bank 22 is a warped filter bank. However, alternatively neither the analysis filter bank 20 nor the OH filter bank 22 are warped.

[0138] Within the onset enhancement unit 56, the first power determination module 24 determines the respective power PA of each of the directed first frequency band signals B’A. Likewise, the second power determination module 26 determines the respective power P’o of each of the directed second frequency band signals Bo, which power P’o is then temporally smoothed by smoother 28 (resulting in a signal referred to as power P”o) and down-sampled to a sampling rate of 2 kHz by down-sampler 30. The signal output by down-sampler 30 is referred to as power Po.

[0139] As mentioned before, the power PA is a time-dependent quantity calculated for each of the 65 center frequencies f\ (i.e. PA = PAC^A)). In contrast, the power Po is a time-dependent quantity calculated for each of the six center frequencies fo (i.e. Po = Po(t,fo)). In both cases, t is a discrete time variable denoting time frames having a frame rate of 2kHz.

[0140] In order to allow comparison of power PA with power Po, the mapping module 32 converts the power PA into a modified signal referred to as power P’A. In the latter, the 65 values of power PA that correspond to the center frequencies f\ of the first frequency band signals BA are reduced to six values corresponding to the center frequencies fo of the second frequency band signals Bo (i.e. P \(t,f \) P’A(t,fo). To this end, the mapping module 32 applies Eq. 4 and a stored mapping function MAO as described above. Thus, each value of the power P’A is derived from a plurality of first frequency band signals BA. The elements of the mapping function MAO are selected such that power P’A has a frequency dependence that corresponds to PA but has the lower frequency resolution of the center frequencies fo.

[0141] The adder 36 (to which the power Po and the negative of the power P’A are fed) compares the (low-delay) power Po with the (larger delay) power P’A and determines the (frequency-dependent) difference of these quantities, which difference is referred to as the comparison result AP. Similar to power Po and power P’A, said comparison result AP is a time- and frequency-dependent quantity with a spectral resolution of the center frequencies fo and the temporal resolution of 2 kHz.

[0142] RestrictedThe comparison result AP is fed to the correction module 38 that calculates a correction gain G’, based on the comparison result AP and the power P’A, which correction gain G’ is a time- and frequency-dependent quantity with a frequency resolution of the center frequencies fo and the temporal resolution of 2 kHz.

[0143] The mapping module 34 converts the correction gain G’ into a modified signal which is referred to as the correction gain G. To this end, the mapping module 34 applies Eq. 1 and a stored mapping function MOA as described above. The elements of the mapping function MOA are selected such that the correction gain G has a frequency dependence that corresponds to the frequency dependence of the correction gain G’ but the finer frequency resolution of the center frequencies f\ (i.e. G’(t,fo) G(t,f\).

[0144] Finally, for each center frequency f\, the multiplier 40 multiplies the correction gain G with the directed first frequency band signals B’A, hereby creating corrected first frequency band signals B”A.

[0145] The corrected first frequency band signals B”A are used as input for one or more signal processing steps. In the exemplary embodiment of Fig. 2, the corrected first frequency band signals B”A are fed to the compressor 42 that applies dynamic compression to said signals, thus creating compressed first frequency band signals B”’A. Thereafter, in the gain handling module 44, the compressed first frequency band signals B”’A are analysed in order to provide a target gain g for each frequency f\. Preferably, said target gain g is adapted to at least partly compensate an individual hearing impairment of the user. It may, thus be calculated based on stored audiogram data of the user. In addition or alternatively, the target gain g may be adapted to at least one of suppress noise and enhance at least one target sound (optionally in dependence of recognition of one of a number of stored sound classes) or to customize the sound to a user preference.

[0146] The filter engine 46 converts the frequency dependent target gain g into time-dependent filter coefficients F’ with a temporal resolution corresponding to the temporal resolution of the analysis filter bank 20 (e.g. 2 kHz). Said filter coefficients F’ are, then, interpolated in time by the interpolation engine 48 to provide the filter coefficients F, e.g. with a temporal resolution of 32 kHz that are provided to the time-domain filter 50.

[0147] RestrictedPreferably, the signal processor 12 is a digital electronic device. In this case, the analog input audio signal I output by the microphones 6 is converted into a digital form before being fed to the spatial analysis module 52 and the analysis filter bank 20, and the signal processor 12 further includes an analog-to-digital converter (not shown) connected upstream of the spatial analysis module 52 and the analysis filter bank 20. Also, a digital-to-analog converter (not shown) is connected - directly or indirectly - downstream of the time domain filter 50 to convert the processed audio signal P or a signal derived therefrom into an analog form.

[0148] The processed audio signal P can be fed directly to the receiver 8 in which case the processed audio signal P is identical to the output audio signal O.

[0149] As an alternative, one or more further signal processing units (not shown), in particular one or more amplifiers and / or an automatic gain control, may be connected downstream of the time-domain filter 50. Moreover, optionally, one or more further signal processing units (not shown), in particular one or more amplifiers are connected upstream of the spatial analysis module 52.

[0150] In the exemplary embodiment of the hearing instrument 2, preferably, the smoother 28 applies a Kalman filter (e.g. in the form of the Kalman smoother previously described) to the power P’o of the second frequency band signals Bo in order to temporally smooth said power P’o. In the concept of the applied Kalman filter (i.e. the Kalman smoother) algorithm, the power of the environment sound is modelled as a stochastic process. Thus, the power P”o reproduced by the smoother 28 is assumed to vary as a consequence of a gaussian noise only:

[0151] P"oJ o = P"oJ - IJo) + WJ - IJo) where W(tJo)~N(Q)

[0152] Eq. 7

[0153] In Eq. 7, W (W = W(t’,fo)) is random variation of P”o within the gaussian distribution of the noise having a variance Q (Q = Q(t’,fo)).

[0154] RestrictedMoreover, the Kalman filter algorithm is based on the further assumption that the measured power P’o fed to the smoother 28 is influenced by a gaussian measurement noise:

[0155] P'o o) = P'o(t' - o) + ^(t' - IJo) where

[0156]

[0157] Eq. 8

[0158] In Eq. 8, V(f,fo) is random error of P’o within the gaussian distribution of the measurement noise having a variance R (R = R(t’,fo)). In Eqs. 7 and 8, similar to the time variable t mentioned above, f is a discrete time variable denoting time frames which, however, have a frame rate of 32 kHz. Thus, t’-l denotes a time frame immediately preceding the current time frame.

[0159] Under the above assumptions, the smother 28 derives the power P”o (f,fo) from

[0160] P

[0161]

[0162] ”o(t'Jo) = P"oW - IJo) + K ■ (P'o(. t' Jo) - P"o(J - rjo))

[0163] Eq. 9

[0164] where the Kalman factor K is given by

[0165] Q(t' ~ W0)

[0166]

[0167] Q(t' - 1, Jo) - R(t' - 1JO)

[0168] Eq. 10

[0169] During operation, the variances Q and R are updated according to the Kalman smoother algorithm as already disclosed and discussed.

[0170] The smoother 28 eliminates or at least reduces temporal fluctuations of the power Po, in particular at low frequencies, that would otherwise occur after down-sampling. Such smoothing was shown to significantly improve the comparison result AP and the correction gain G derived therefrom.

[0171] In alternative embodiments of the hearing instrument 2, the Kalman filter described above is replaced by sliding or recursive averaging or by normalized least-mean-square

[0172] Restrictedaveraging. It was shown that those averaging methods have a similar effect as the Kalman filter described above. However, they add a certain delay to the power Po and are, thus counter-productive to the aim of low-delay onset enhancement. Therefore, implementing one of these averaging methods in the smoother 28, instead of the Kalman filter, is less preferred but still within the scope of the invention.

[0173] In the exemplary embodiment of the hearing instrument 2, preferably, the ensemble of the correction module 38, the adder 36 and the mapping modules 32 and 34 apply Eqs. 1 to 4 to the power Po and the power PA to derive the correction gain G. Herein, preferably, the correction module 38 uses a sharp step function as shown in Eq. 6 as the correction function C.

[0174] It will be appreciated by persons skilled in the art that numerous variations and / or modifications may be made to the invention as shown in the specific examples without departing from the spirit and scope of the invention as broadly described in the claims. The present examples are, therefore, to be considered in all aspects as illustrative and not restrictive.

[0175] RestrictedLIST OF REFERENCE NUMERALS

[0176] 2 hearing instrument

[0177] 4 housing

[0178] 6 microphone

[0179] 8 receiver

[0180] 10 battery

[0181] 12 signal processor

[0182] 14 sound channel

[0183] 16 tip

[0184] 20 (analysis) filter bank

[0185] 22 (onset enhancement) filter bank 24 power determination module 26 power determination module 28 smoother

[0186] 30 down- sampler

[0187] 32 mapper

[0188] 34 mapper

[0189] 36 adder

[0190] 38 correction module

[0191] 40 multiplier

[0192] 42 compressor

[0193] 44 gain handling module

[0194] 46 filter engine

[0195] 48 interpolation engine

[0196] 50 time domain filter

[0197] 52 spatial analysis module

[0198] 54 spatial analysis module

[0199] 56 onset enhancement module

[0200] ΔP comparison result

[0201] g target gains

[0202] BA (first) frequency band signals

[0203] RestrictedB’A (directed first) frequency band signals B”A (corrected first) frequency band signals B”’A (processed first) frequency band signals Bo (second) frequency band signals

[0204] F filter coefficients

[0205] F’ filter coefficients

[0206] G correction gain

[0207] G’ correction gain

[0208] I input audio signal

[0209] I' directed (time-domain) audio signal MAO(first) mapping function

[0210] MOA(second) mapping function

[0211] NAnumber (of first frequency band signals)NOnumber (of second frequency band signals)O output audio signal

[0212] P processed audio signal

[0213] PA power

[0214] P’A power

[0215] Po power

[0216] P’o power

[0217] P”o power

[0218] U voltage

[0219] Restricted

Claims

CLAIMS1. A method for operating a hearing instrument (2), the method comprising:- capturing a sound signal (I) from an environment of the hearing instrument (2);- providing a processed sound signal (P) by processing said captured sound signal (I) or a signal (L) derived therefrom;- outputting said processed sound signal (P) or a signal (O) derived therefrom to a user of the hearing instrument (2);wherein the method further comprises:- splitting, by a first filter bank (20), said captured sound signal (I) or a signal (L) derived therefrom into a number (NA) of first frequency band signals (BA);- applying at least one processing step to the first frequency band signals (BA) or signals (B’A, B”A) derived therefrom to contribute to said providing of said processed sound signal (P);- splitting, by a second filter bank (22), said captured sound signal (I) or a signal (L) derived therefrom into a number (No) of second frequency band signals (Bo), wherein each of the second frequency band signals (Bo) has a lower delay vis-a-vis said captured sound signal (I) as at least one spectrally matching first frequency band signal (BA);- comparing each one of at least a subset of the second frequency band signals (Bo) or a signal (Po) derived therefrom with at least one spectrally matching first frequency band signal (BA), or with a signal (P’A) derived therefrom, to receive a comparison result (AP);- calculating a correction gain (G) based on the comparison result (AP);- applying said correction gain (G) to the first frequency band signals (BA) or signals (B’A) derived therefrom to provide corrected first frequency band signals (B”A); and- using said corrected first frequency band signals (B”A) to at least partly control said processing of said captured sound signal (I) or signal (L) derivedRestrictedtherefrom.

2. The method according to claim 1 wherein said processing of said captured sound signal (I) or signal (I’) derived therefrom comprises filtering said captured sound signal (I) or signal (I’) derived therefrom in a time domain digital filter, wherein said digital filter is either a low delay or minimum phase digital filter.

3. The method according to claim 1, wherein the second filter bank (22) comprises a window function that is more narrow than the window function comprised in the first filter bank (20).

4. The method according to any of the preceding claims, wherein said first filter bank (20) and said second filter bank (22) are adapted to provide the same number of frequency band signals.

5. The method according to any of the claims 1-3, wherein the number (NA) of first frequency band signals (BA) exceeds the number (No) of second frequency band signals (Bo) or exceeds by a factor of at least 4 or by a factor of at least 8 or by a factor of at least 10.

6. The method according to any of the preceding claims, wherein the respective power (Po) of each one of at least a subset of the second frequency band signals (Bo) or a signal (Po) derived therefrom is smoothed using a Kalman-Filter, prior to the comparing step.

7. The method according to any of the preceding claims, wherein the correction gain (G) is determined according to the formulas:= ^mOA(fA,fo) ■foandRestrictedP'_A(t,f_O) = Σ m_AO(f_O,f_A) · P_A(t,f_A)fAwherein- G is the correction gain,- PA is a power derived from the first frequency band signal (BA) at frequency fA,- Po is a power derived from the second frequency band signal (Bo) at frequency fo,- mAo is an element of a first mapping function MAO that maps the power (PA) of the first frequency band signals (BA) to the frequencies (fo) of the second frequency band signals (Bo), and- moA is an element of a second mapping function MOA that maps correction gains (G’) calculated for frequencies fo of the second frequency band signals (Bo) to the frequencies \ of the first frequency band signals (BA), - C is an optional correction function to be multiplied to the comparison result Po - P’A; and- t is time.

8. A hearing instrument (2) comprising:- an input transducer (6) configured to capture a sound signal (I) from an environment of the hearing instrument (2);- a first filter bank (20) configured to split said captured sound signal (I) or a signal (I’) derived therefrom into a number (NA) of first frequency band signals (BA);- an output transducer (8) configured to output a processed sound signal (P) or a signal (O) derived therefrom to a user of the hearing instrument (2);- a signal processor (12) configured to apply at least one processing step to the first frequency band signals (BA) or signals (B’A, B”A) derived therefrom to contribute to providing said processed sound signal (P) or a signal (O) derived therefrom; andRestrictedthe hearing instrument (2) further comprising:- a second filter bank (22) configured to split the captured sound signal (I) or a signal (I’) derived therefrom into a number (No) of second frequency band signals (Bo), wherein each of the second frequency band signals (Bo) has a lower delay vis-a-vis the captured sound signal (I) as at least one spectrally matching first frequency band signal (BA);- an onset enhancement module (56) configured to:- - compare, in a comparison step, each one of at least a subset of the second frequency bands signals (Bo) or a signal (Po) derived therefrom with at least one spectrally matching first frequency band signal (BA) or a signal (P’A) derived from a plurality of spectrally matching first frequency band signals (BA) to receive a comparison result (AP);- - calculate a correction gain (G) based on the comparison result (AP); and - - apply said correction gain (G) to the first frequency band signals (BA) or signals (B’A) derived therefrom to provide corrected first frequency band signals (B”A); and- - use said corrected first frequency band signals (B”A) to at least partly control said processing of said captured sound signal (I) or signal (I’) derived therefrom.

9. The hearing instrument according to claim 8, wherein the number (NA) of first frequency band signals (BA) exceeds the number (No) of second frequency band signals (Bo) or exceeds by a factor of at least 4 or by a factor of at least 8 or by a factor of at least 10.Restricted

Images