EP1853089A2 - Method for elimination of feedback and for spectral expansion in hearing aids - Google Patents

Abstract

The method involves determining a feedback-endangered frequency range. An input signal is received with a spectral portion in the frequency range. The known spectral portion of the input signal is reduced. The reduced spectral portion is mixed with a synthetic signal, in such a manner that in the spectral region, the efficiency of the entire signal corresponds to the efficiency before the reduction. The synthetic signal is produced by frequency shift from the input signal. A spectral envelope of the mixed signal is corrected with the help of linear predictive coding analysis. Independent claims are also included for the following: (1) a hearing device with feedback suppression device (2) a method for providing the spectral expansion in a hearing device.

Description

The present invention relates to a method for suppressing feedback whistles in hearing devices by determining or predetermining a frequency range that is subject to feedback risk, and receiving an input signal having a spectral component in the feedback-prone frequency range. Moreover, the present invention relates to a method for spectral expansion in a hearing device whose input signal has a limited frequency range. In addition, the present invention relates to corresponding hearing devices.

In acoustic systems and in particular in hearing aids with at least one input (eg microphone) and at least one output (eg listener) there is a risk of acoustic feedback. If the gain is sufficiently high, the system starts to oscillate, which is noticeable by whistling.

So far, the feedback whistles could be suppressed for example by so-called notch filter. In this approach, the loop gain is reduced at the frequency at which feedback or feedback whistles would occur. Due to this reduction, the amplitude condition for feedback whistles is no longer met.

Another way to suppress the feedback whistles is to perform a corresponding signal compensation. This feedback compensation approach digitally replicates the feedback path and compensates for its effect. However, these approaches to feedback reduction can significantly audibly distort the output signal, especially if the input stage of the acoustic system is designed for only a low spectral bandwidth.

Acoustic systems with narrowband input stage also have the disadvantage that the acoustic quality of the output signal is usually correspondingly low.

From the publication EP 1 304 902 A1 For example, a method and apparatus for noise cancellation of a redundant acoustic signal is known. In this case, a sub-frequency range of the input signal in which a disturbance is concentrated is removed. Subsequently, the intensity of the remaining input signal is divided into an input signal part to be maintained and an input signal part to be further processed. Due to the further processed input signal part of the remote sub-frequency range of the input signal is synthesized. Finally, the input signal portion to be retained and the synthesized input signal portion are merged to produce an output reduced in interference with the input signal.

The object of the present invention is therefore to improve the signal quality of acoustic systems which are subject to feedback or whose input stage is relatively narrow-band.

According to the invention, this object is achieved by a method for suppressing feedback whistling in a hearing device by determining or predetermining a frequency range that is subject to feedback, and receiving an input signal with a spectral component in the feedback-prone frequency range, and reducing said spectral component of the input signal and mixing the reduced spectral component with a synthetic signal, so that in the said spectral range the power of the total signal substantially corresponds to the power before reducing.

Accordingly, the invention provides a hearing device with a feedback suppression device and a signal input device for receiving an input signal, wherein the feedback suppression means comprises a reduction unit for reducing a spectral component of the input signal and a mixing unit for mixing the reduced spectral component with a synthetic signal, such that in said spectral range the power of the overall signal substantially corresponds to the power before reduction.

The invention is based on the idea to substitute a part of an internal signal of the hearing by a synthetic signal and to mix with this. By substituting, the amplitude condition for the feedback whistling is no longer satisfied.

Preferably, the synthetic signal is generated with a nonlinearity from the input signal. In this way, in the desired frequency range, a synthetic signal in response to the input signal can be generated.

For example, the synthetic signal may also be generated by frequency shifting from the input signal. This also makes it possible to generate a synthetic signal as a function of the input signal in a simple manner in the desired frequency range.

Advantageously, the spectral envelope of a signal mixed from the synthetic signal and a part of the input signal is corrected by means of LPC analysis. Thus, the signal character of the original input signal can be well maintained without feedback. For example, the correction can be done in combination with a common shape filtering.

According to a particular embodiment of the present invention, prior to mixing, further processing of the reduced signal and mixing is performed by adding the synthetic signal to the further processed, reduced signal just prior to signal output to an output transducer. Thus, the suppression of the feedback whistle can be completely independent of the internal signal processing. This means that existing systems can also be easily retrofitted.

Furthermore, the input signal can be processed in a plurality of channels, wherein the substitution or mixing takes place only in that channel with the backward-endangered frequency range. Thus, the effect of the feedback suppression can be selectively limited to one or more channels. It is advantageous if one or more features of the respective signal are obtained from at least two of the channels and considered for substitution or mixing. On the basis of the characteristics from the other channels, the quality of the synthetic signal can be improved.

In order to achieve the above object, there is further provided a method for spectral expansion in a hearing apparatus by receiving an input signal whose spectrum has a limited a priori a limited frequency range and mixing the input signal or the input signal in a further processed form with a synthetic signal whose spectrum is at least partially outside the limited frequency range.

In addition, the invention provides a corresponding hearing device with a signal input device for receiving an input signal whose spectrum a priori has a limited frequency range and a mixer for mixing the input signal or the input signal in a further processed form with a synthetic signal whose spectrum is at least partially outside the limited Frequency range is.

The inventive mixing of the input signal with a synthetic signal, a spectral expansion is achieved, which leads to an output signal, which is perceived as a higher quality. This is the spectral expansion achieved by a relatively small amount of hardware. In addition, in certain acoustic systems in which technically conditioned limitations of the bandwidth of the input stage exist, the inventive spectral extension can be used to the extent that the bandwidth is not limited in the output signal.

According to a preferred embodiment, the synthetic signal is generated by copying a portion of the limited frequency range of the input signal. Specifically, mirror frequencies can be used when copying. Thus, an input signal dependency of the synthetic signal can be easily generated.

According to a further embodiment of the system or method according to the invention, the mixing of the input signal with the synthetic signal can be interrupted if a non-linear behavior of the hearing device is detected. In this way, a noisy feedback signal can be prevented, which would not tear off by itself.

FIG 1

ein Prinzipschaltbild einer Hörvorrichtung gemäß einer ersten Ausführungsform der vorliegenden Erfindung und

FIG 2

ein Prinzipschaltbild zur erfindungsgemäßen Teilbandsynthese eines Mehrkanalgeräts.

The present invention will now be explained in more detail with reference to the accompanying drawings, in which:

FIG. 1

a schematic diagram of a hearing aid according to a first embodiment of the present invention and

FIG. 2

a schematic diagram of the invention subband synthesis of a multi-channel device.

The embodiments described in more detail below represent preferred embodiments of the present invention.

According to the basic idea according to the invention, signal components which cause the feedback whistling are to be substituted become. This signal substitution should take place in the feedback-endangered frequency range. In this frequency range, therefore, not only the signal picked up by the microphone is processed and delivered via the receiver, but also the synthetically generated signal is processed or output. Thus, the feedback loop can be interrupted and it can be prevented in case of linear system behavior unwanted oscillation.

The signal picked up by the microphone can be mixed with the synthetic signal in any ratio. This mixing can also be considered as a partial substitution. In this case, the effective amplification can be lowered to such an extent in the feedback loop that the amplitude condition for feedback is no longer satisfied. As a result, a certain proportion of the natural signal is retained.

Measures for generating synthetic signal components include, for example, the use of non-linearities, i. H. non-linear components with, for example, quadratic characteristic, magnitude characteristic, etc., or modulation approaches in which frequency components are spectrally shifted. In addition, especially in low frequency position (<8 kHz), a device for correcting the spectral envelope should be provided in order to preserve a natural sound as much as possible. One tool for this is, for example, the LPC analysis (linear predictive coding) in combination with shape filtering.

In the case of the suppression of feedback whistles according to the invention, it is advantageously sufficient to know in which frequency band feedback whistling occurs or can occur. The nominal power in the relevant frequency band is not reduced as in the notch filter approach. Rather, virtually no power is lost in the signal substitution according to the invention in that frequency band in which feedback whistling occurs. In addition, the feedback path must be not explicitly known in the inventive solution, as is necessary in the feedback compensation approach.

In Figure 1, a concrete implementation example is presented. In a switch 1, the original input signal of a microphone 2 is split into two complementary spectral ranges. In the present case, the switch 1 contains a band-stop filter 3 and a band-pass filter 4. As a result, the signal is divided into a band-pass signal S_fb and a spectrally complementary signal S_kompl. Instead of bandpass filtering, low-pass or high-pass filtering can also be used.

The spectral range of the bandpass signal S_fb represents the band in which feedback whistles would arise without countermeasures. The bandpass signal S_fb is multiplied by a factor a in a multiplier 5. Multiplied by this factor a (where 0 <a <1), the bandpass signal S_fb is partially added back to the complementary signal S_kompl in the adder 6. The signal thus obtained passes through the regular signal processing 7, which would pass through the original signal even without compensation measure for feedback whistles.

The output signal of the microphone 2 is also used to generate the synthetic signal in the spectral range of the bandpass signal S_fb corresponding to the lower path of FIG. For example, a suitable spectral band is cut out by means of a filter and copied into the spectral band of interest. Corresponding means for generating a synthetic signal 8 are shown in the lower path of the circuit diagram of FIG. The synthetic signal is weighted by a factor b. This weighting with the aid of a multiplier 9 can take place prior to entry into the means for generating the synthetic signal 8. Subsequently, the synthetic signal is adjusted by means of a signal processing module 10 so that it can be added to the signal of the signal processing 7 of the upper path. This addition takes place in an adder 11 immediately before the signal output to an output converter, not shown in FIG 1.

The factors a and b are coordinated. They define the mixing ratio of synthetic and real signal component in the spectral range of the bandpass signal S_fb. The larger the factor a, the smaller must be the factor b and vice versa, so that the feedback whistling can be suppressed. In a first extreme case, a is close to 1 and b is close to 0, so that practically no signal substitution by a synthetic signal takes place in the spectral range of the bandpass signal S_fb. In a second extreme case, a is close to 0 and b is close to 1, which results in almost complete signal substitution by the synthetic signal in the spectral range of the bandpass signal S_fb.

According to a development of the embodiment of FIG. 1, features of the original signal can be extracted from the signals of the upper path. With these features, a correction of the spectral envelope in the synthesized band can be achieved. FIG. 2 shows a circuit diagram of a multi-channel device with subband synthesis and feature extraction. The output signal of a microphone 20 is again decomposed into two channels. For this purpose, the first filter is, for example, a high-pass filter 21 and the second filter is a low-pass filter 22. The high-pass signal corresponds to a channel A and the low-pass signal corresponds to a channel B. A hearing aid signal processing unit 23 is arranged in the channel A and a hearing aid signal processing unit 24 in the channel B. , The output signals of the two signal processing units 23 and 24 are added in an adder 25 and sent the sum signal to a handset 26.

A part of the acoustic output signal of the handset 26 is fed back to the microphone 20 via a feedback path 27. Since the feedback takes place primarily in the high-frequency channel A, is between the high-pass filter 21 and the hearing aid signal processing unit 23 a mixer 28 connected, with which a synthetic signal can be mixed in the high-frequency channel. To generate the synthetic signal, one or more features of the high-frequency channel A are obtained by a feature extraction unit 29 and also one or more features of the low-frequency channel B by a feature extraction unit 30. The features obtained by the units 29 and 30 are evaluated or compared in an evaluation unit 31. The evaluation unit 31 is based on a model 32. This model includes a prior knowledge of ratios of high-pass to low-pass shares. The evaluation unit 31 thus determines, for example, based on the spectral envelope, which is available as a feature from the high-frequency channel A, and the model 32, a mixing ratio for the mixing stage 28. In addition, the evaluation unit 31 controls a signal generator 33, for. As a vocoder. The signal generator 33 then supplies the synthetic signal to the mixer 28.

The example of FIG. 2 shows a two-channel hearing device. However, the invention can also be applied to any other devices with two or more channels.

The above-mentioned mixing or substitution can also be used for spectral broadening. For example, in an acoustic system having at least one input (eg microphone, receiver) and at least one output (eg listener) one or more frequency ranges of the signal to be output are to be generated synthetically. Thus, the input stage of the acoustic system can be designed for a lower spectral bandwidth or in systems whose input stage can not technically exceed a certain bandwidth, it is possible to extend the bandwidth of the output signal to a larger target bandwidth. The advantage of this is that the spectral expansion is possible by a relatively small amount of hardware. In addition, there are technical limitations the bandwidth of the input stage does not include the bandwidth of the output signal.

As a concrete example here is a wireless audio connection (wireless audiolink) called. The limiting element in the input stage is the receiver, which delivers a maximum of 8 kHz. Since frequencies up to 12 kHz are required in hi-fi operation, the band is synthesized from 8 kHz to 12 kHz.

Another implementation variant for the spectral expansion according to the invention relates to hearing aids. The synthetic generation of spectral components above 8 kHz is very advantageous for hearing aids, since above this frequency is to be feared feedback whistles. Even without correction of the spectral envelope, a clear spectral expansion can be perceived by copying lower frequency bands into the band above 8 kHz. For example, the utilization of image frequencies outside of the Nyquist band can serve as the copying process, and the "by-products" of frequency shift operations are purposefully exploited.

Even if a frequency band is occupied by synthetic spectral components and thus no feedback whistling in the traditional sense (oscillation of an unstable, linearly time-invariant system) can arise in this band, there is nevertheless a comparable feedback phenomenon with correspondingly high amplification. In fact, real systems behave non-linearly, especially at the driving limit. The reason for this is, for example, the non-linear behavior of hardware components, eg. B. handset or microphone, but also nonlinearities in digital signal processing, z. Hard limiters or AGCs. If the synthetic spectral components are derived from natural spectral components outside of the frequency band to be occupied, a closed feedback loop may arise as follows: a natural spectral component becomes a synthetic one according to a given algorithm Spectral component generated; interfering non-linearity, in turn, generates out-of-band spectral components with synthetic spectral components; the newly generated spectral components are fed back to the microphone; The newly generated spectral components again serve as the basis for the generation of synthetic spectral components, whereby the loop is closed. In extreme cases, this creates a noisy feedback signal that does not break off on its own.

However, a remedy against the noise-like feedback signal can be provided that the non-linear behavior of the system is detected, for example by overdrive detection. If the system behaves non-linearly for some time (eg, oversteer), the synthetic generation is momentarily interrupted (eg, <1 second) so that the self-stabilized feedback noise can be broken off.

Claims (16)

A method for suppressing feedback whistling in a hearing device Determining or presetting a frequency range that is vulnerable to feedback, and Receiving an input signal having a spectral component in the feedback-prone frequency range, characterized by - Reduce said spectral component of the input signal and - Mixing (11) of the reduced spectral component with a synthetic signal (8), so that in the said spectral range the power of the total signal substantially corresponds to the power before reducing. The method of claim 1, wherein the synthetic signal having a nonlinearity is generated from the input signal. The method of claim 1 or 2, wherein the synthetic signal is generated by frequency shifting from the input signal. Method according to one of the preceding claims, wherein the spectral envelope of the mixed signal is corrected by means of an LPC analysis. The method of claim 4, wherein the correction is in combination with shape filtering. A method according to any one of the preceding claims, wherein prior to mixing, further processing of the reduced signal and mixing is performed by adding the synthetic signal to the further processed, reduced signal just prior to signal output to an output transducer. Method according to one of the preceding claims, wherein the input signal in a plurality of channels (A, B) is processed and the mixing takes place only in the channel with the feedback-vulnerable frequency range. The method of claim 7, wherein from at least two of the channels (A, B) each one or more features of the respective signal (29, 30) obtained and taken into account for the mixing. Hearing device with a feedback suppression device and a signal input device for receiving an input signal, characterized in that the feedback suppression device comprises a reduction unit (5) for reducing a spectral component of the input signal and - Has a mixing unit (11) for mixing the reduced spectral component with a synthetic signal, so that in said spectral range, the power of the total signal substantially equal to the power before reducing. A method for spectral expansion in a hearing by Receiving an input signal whose spectrum has a priori a limited frequency range and Mixing the input signal or the input signal in a further processed form with a synthetic signal whose spectrum is at least partially outside the limited frequency range. The method of claim 10, wherein the synthetic signal is generated by copying the fraction from the limited frequency range of the input signal. A method according to claim 10 or 11, wherein image frequencies are used during copying. A method according to any one of claims 10 to 12, wherein the mixing of the input signal with the synthetic signal is interrupted when a non-linear behavior of the hearing device is detected. A method according to any one of claims 10 to 13, wherein the spectral envelope of the mixed signal is corrected by LPC analysis. The method of any one of claims 10 to 14, wherein the mixing with the synthetic signal occurs immediately prior to signal output to an output transducer. Hearing device with - A signal input device for receiving an input signal whose spectrum a priori has a limited frequency range and - A mixing device (11) for mixing the input signal or the input signal in a further processed form with a synthetic signal whose spectrum is at least partially outside the limited frequency range.

Images