EP2109329A2 - Multi-layer estimation method for reducing interference noises and audio device - Google Patents

Abstract

The method involves estimating a value e.g. signal-to-noise ratio, of an information signal (S) with an estimation algorithm by an estimating device (NS1), and parameterizing another estimation algorithm with the estimated value. Another value of the signal is estimated with the latter algorithm by another estimating device (NS2). Noise in the signal is controlled based on the latter estimated value by a noise reduction device (NR). A temporal change rate of the signal is estimated as a value for the parameterizing of the latter algorithm by the former algorithm. An independent claim is also included for a hearing device comprising an estimating device for estimating a value of an input signal.

Description

The present invention relates to a method for noise reduction for hearing aids by estimating a value of an input signal with an estimation algorithm. Moreover, the present invention relates to a corresponding hearing apparatus having estimating means for estimating a value of an input signal with an estimation algorithm and a noise reduction means for reducing a noise in the input signal. The term "hearing device" is understood here to mean any sound-emitting device which can be worn in or on the ear, in particular a hearing device, a headset, headphones and the like.

Hearing aids are portable hearing aids that are used to care for the hearing impaired. In order to meet the numerous individual needs, different types of hearing aids such as behind-the-ear hearing aids (BTE), hearing aid with external receiver (RIC: receiver in the canal) and in-the-ear hearing aids (IDO), e.g. Concha hearing aids or canal hearing aids (ITE, CIC). The hearing aids listed by way of example are worn on the outer ear or in the ear canal. In addition, bone conduction hearing aids, implantable or vibrotactile hearing aids are also available on the market. The stimulation of the damaged hearing takes place either mechanically or electrically.

Hearing aids have in principle as essential components an input transducer, an amplifier and an output transducer. The input transducer is usually a sound receiver, z. As a microphone, and / or an electromagnetic receiver, for. B. an induction coil. The output transducer is usually used as an electroacoustic transducer, z. As miniature speaker, or as an electromechanical transducer, z. B. bone conduction, realized. The amplifier is standard integrated into a signal processing unit. This basic structure is in FIG. 1 shown using the example of a behind-the-ear hearing aid. In a hearing aid housing 1 for carrying behind the ear, one or more microphones 2 for receiving the sound from the environment are installed. A signal processing unit 3, which is also integrated in the hearing aid housing 1, processes the microphone signals and amplifies them. The output signal of the signal processing unit 3 is transmitted to a loudspeaker or earpiece 4, which outputs an acoustic signal. The sound is optionally transmitted via a sound tube, which is fixed with an earmold in the ear canal, to the eardrum of the device carrier. The power supply of the hearing device and in particular the signal processing unit 3 is effected by a likewise integrated into the hearing aid housing 1 battery. 5

When processing digital voice recordings, eg. As digital hearing aids, it is often desirable to suppress disturbing background noise without affecting the useful signal (voice). For this purpose, filtering methods which influence the short-term spectrum of the signal, such as the Wiener filter, are known and suitable. However, these methods require accurate estimation of the frequency dependent power of the noise to be canceled from an input signal. If this estimate is inaccurate, either an unsatisfactory noise suppression is achieved, the desired signal is attacked or there are additional artificially generated noise, so-called "musical tones". Noise estimation methods that fully and efficiently solve these problems are not yet available.

1. 1. Sprachaktivitätserkennung:
   • Solange keine Sprachaktivität festgestellt wird, betrachtet man die komplette (zeitveränderliche) Eingangssignalleistung als Störgeräusch. Sofern Sprachaktivität detektiert wird, hält man die Störgeräuschschätzung auf dem letzten, vor dem Einsetzen der Sprachaktivität geschätzten, Wert konstant.
2. 2. Störleistungsschätzung während einer Sprachaktivität (so genanntes "Minimum-Tracking-Verfahren"):
   • Es ist bekannt, dass auch während einer Sprachaktivität die Sprachsignalleistung in einzelnen Frequenzbereichen immer wieder kurzfristig nahezu Null ist. Liegt nun eine Mischung aus Sprache und vergleichsweise langsam zeitveränderlichem Störgeräusch zugrunde, so entsprechen die Minima der zeitlich betrachteten spektralen Signalleistung der Störgeräuschleistung zu diesen Zeitpunkten. Zwischen den festgestellten Minima muss die Störsignalleistung liegen ("Minimum-Tracking"). Ein derartiges Minimum-Tracking kann beispielsweise mit Hilfe eines Glättungsfilters durchgeführt werden, der beispielsweise beschrieben ist in R. Martin, "Noise power spectral density estimation based on optimal smoothing and minimum statistics", IEEE Trans. Speech Audio Processing, Vol. 9, Nr. 5, Juli 2001, Seiten 504 - 512 oder S. Rangachari, P. Loizou, "A noise-estimation algorithm for highly non-stationary environments", Speech Communication, Vol. 48, Februar 2006, Seiten 220 - 231 . Die Ermittlung der Störgeräuschleistung erfolgt typischerweise getrennt für verschiedene Frequenzbereiche im Eingangssignal. Hierzu wird das Eingangssignal zunächst mittels einer Filterbank oder einer Fourier-Transformation in einzelne Frequenzkomponenten aufgespalten. Diese Komponenten werden dann getrennt voneinander verarbeitet.

So far, the noise power can be estimated in principle by two approaches. Both methods can take place either broadband or preferably in a frequency domain decomposition by means of filter bank or short-time Fourier transformation:

1. 1. Voice Activity Detection:
   • As long as no voice activity is detected, the complete (time-varying) input signal power is regarded as noise. If speech activity is detected, the noise estimate is kept constant at the last value estimated before the onset of speech activity.
2. 2. Disturbance power estimation during a voice activity (so-called "minimum tracking method"):
   • It is known that even during a voice activity the voice signal power in individual frequency ranges is almost always close to zero in the short term. If a mixture of speech and comparatively slowly time-varying noise is now the basis, the minima of the time-considered spectral signal power correspond to the noise power at these times. The interference signal power must lie between the detected minimums ("minimum tracking"). Such minimum tracking can be carried out, for example, with the aid of a smoothing filter, which is described, for example, in US Pat R. Martin, "Noise power spectral density estimation based on optimal smoothing and minimum statistics", IEEE Trans. Speech Audio Processing, Vol. 9, No. 5, July 2001, pages 504-512 or S. Rangachari, P. Loizou, "A noise-estimation algorithm for highly non-stationary environments," Speech Communication, Vol. 48, Feb. 2006, pages 220-231 , The determination of the noise power is typically done separately for different frequency ranges in the input signal. For this purpose, the input signal is first split into individual frequency components by means of a filter bank or a Fourier transformation. These components are then processed separately.

In the above-mentioned method 1, on the one hand, reliable recognition of voice activity is a problem, and on the other hand, it is not possible to change time Track noise during simultaneous speech activity.

In the above method 2 fundamental contradictions in the setting of the algorithm to be solved: If speech is present, the noise estimate should be adapted only slowly, so as not to classify speech components as noise by rapid adaptation and thereby attacking the speech quality. If no speech is present, then the noise power estimation should follow without delay the temporal fine structure of the input signal. This results for the adjustment parameters of the method, such. B. smoothing time constants, window length for a minimum search or weighting factors contradictory requirement that could be optimally solved so far only on average.

The object of the present invention is therefore to improve the quality of a noise suppression, so that in particular language is less attacked and disturbing artifacts are better avoided.

According to the invention, this object is achieved by a method for noise reduction for hearing devices by estimating a first value of an input signal with a first estimation algorithm, parameterizing a second estimation algorithm with the estimated first value, estimating a second value of the input signal with the second estimation algorithm, and reducing noise in the first estimation algorithm Input signal based on the estimated second value. Here, the first value can be equal to the second value here as in the following.

Moreover, according to the invention, a hearing device and a hearing device are provided with a first estimation device for estimating a first value of an input signal with a first estimation algorithm and a noise reduction device for reducing a noise in the input signal and comprising a second estimation device that is parameterized with the estimated first value , to the Estimating a second value of the input signal with a second estimation algorithm, wherein the noise reduction means obtains the estimated second value from the second estimation means for reducing the noise.

The two-stage estimation according to the invention leads to a significantly improved estimation quality, because in the first stage a simple estimation can be carried out, the result of which is used for the parameterization of the second estimation device or of the second estimation algorithm. The second estimation algorithm can thus be adapted to a specific interference situation, as a result of which a situation-specific estimation can be achieved.

The first estimation algorithm may be based on a minimum tracking method. This can be determined in a simple way an interference power level in voice activity.

In a specific embodiment, the first estimation algorithm can estimate a temporal rate of change of the input signal as the first or further value for the parameterization of the second estimation algorithm. Thus, the total power and the interference power can be reliably estimated.

According to another embodiment, the first estimation algorithm and the second estimation algorithm may be structurally equal. This reduces the implementation effort. In particular, it is thus possible for the first estimation device and the second estimation device to be realized by a single estimation device, which is operated alternately as a first and a second estimation device in time division multiplex.

The two estimation algorithms can also be different. Thus, the first estimation algorithm may include a recursive smoothing and the second estimation algorithm may not be recursive. In this way, the implementation effort can be adapted to the desired estimation quality.

Preferably, the first value estimated by the first estimation algorithm is a signal power, an interference power, or a signal-to-interference ratio. These quantities can be used directly to dampen corresponding interferences.

Furthermore, a first value can be selectively estimated by the first estimation algorithm for a plurality of frequency ranges and these first values can be combined to parameterize the second estimation algorithm. This makes it possible to influence the parameterization of the second estimation algorithm based on the spectral distribution of the input signal.

Particularly preferred is the dynamic parameterization of the second estimation algorithm with a constantly updated first value of the first estimation algorithm. Thus, the noise reduction constantly adjusted to the current acoustic situation can always be done with high quality.

Furthermore, it may be favorable in the described method for noise reduction to split the input signal into individual frequency components and to process them with respect to the un-split signal in time sub-sampled form. With this downward scan the computing effort can be significantly reduced.

FIG 1

den prinzipielle Aufbau eines Hörgeräts gemäß dem Stand der Technik und

FIG 2

ein Blockdiagramm einer Realisierungsform eines er- findungsgemäßen Verfahrens.

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

FIG. 1

the basic structure of a hearing aid according to the prior art and

FIG. 2

a block diagram of an embodiment of a method according to the invention.

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

In the FIG. 2 illustrated signal processing device of a hearing aid has an Anlaysefilterbank AFB at the signal input. It has a broadband signal input BI and a multi-channel output CO. In the broadband input BI, a disturbed useful signal S is fed. This signal is spectrally decomposed by the analysis filter bank AFB. The output signal of the analysis filter bank AFB is fed to the input I1 of a first estimator NS1, to an input 12 of a second estimator NS2 and to an input 13 of a noise reduction device NR. The first estimator NS1 estimates the power of the interfering signal and outputs it as an initial disturbance power at the output NP1. In addition, the estimator NS1 here also estimates the useful signal power and outputs it at the output SP1.

The second estimator NS2 receives, in addition to the output signal of the analysis filter bank AFB, the initial interference signal power at its input NP2 and the initial useful signal power at its input SP2. The initial noise power and the initial useful signal power are used to parameterize the adaptive estimator NS2. With the current parameter setting, the second estimator estimates a final noise signal power that it outputs at its output FNP2 and, optionally, a final useful signal power that it outputs at its output FSP2.

The second signal downstream of the adaptive second estimator NS2, which may be implemented as a Wiener filter, for example, receives the final interference signal power at its input FNP3 and the final useful signal power at its input FSP3. Based on these quantities together with the output signal of the analysis filter bank AFB, the noise reduction algorithm calculates the noise reduction device NR an attenuation or reduction gain, which is output at the output RG.

The preferred multi-channel reduction gain of the noise reduction device NR is supplied together with the multi-channel output of the analysis filter bank a multiplier M, which performs a multiplication channel by channel, so that a multi-channel noise-free signal is formed, which is fed to a synthesis filter bank SFB specifically their multi-channel input CI. The synthesis filter bank SFB synthesizes the signals of the individual channels into a broadband noise-reduced output signal SR. This signal is available at the BO output.

The noise reduction is thus based on a two-stage estimate of the noise power. In this case, a first estimate of the total power or the useful signal power and the interference power in the first estimator NS1 is initially carried out. This first estimation can be done, for example, by means of a permanently parameterized minimum tracking method, as described above. For example, the temporal rate of change of the input signal can also be used as an (possibly additional) criterion for the estimation. This rate of change is in the Essay FF Quatieri, RB Dunn, "Speech enhancement based on auditory spectral change", Proc. IEEE Int. Conf. Acoustics, Speech, Signal Processing (ICASSP), Vol. I, 2002, pages 257 to 260 described under the keyword "spectral change".

Because of this estimate of the first estimator NS1, e.g. For example, in the form of a signal-to-noise ratio (or in a preferred embodiment in the form of a noise signal ratio and / or noise power directly, operating parameters of the second, parallel to the first operated disturbance estimation method are adapted in the second estimator NS2.

In a preferred embodiment, the second interference estimation method is structurally the same as the first one and differs only by the parameterization which has been changed adaptively on the basis of the results of the first method. In the second estimator, for example, a time constant of a smoother can be adapted so that a faster smoothing occurs at a low estimated signal-to-noise ratio than at a high estimated signal-to-noise ratio.

In the second estimator NS2, furthermore, not only one parameter but also several parameters can be changed on the basis of the estimated values from the first estimator. The change of the parameters of the second disturbance power estimator NS2 can be frequency-dependent directly according to the first estimate of the disturbance power. Alternatively, the change in the parameters of the second disturbance power estimator can also be made on the basis of a summary of the originally frequency-selectively determined first noise estimate. In this case, the change ranges and limit values of the parameters of the second noise estimator NS2 can be determined frequency-dependent. In particular, it is advantageous if the change ranges and limit values of the second noise estimator NS2 are set dynamically as a function of the first estimate.

The second noise estimation method and the second noise estimator may also be structurally different from the first one. So z. For example, recursive smoothing (see R. Martin, supra) may be used in the first method, while in the second, a non-recursive method (see S. Rangachari, P. Loizou, supra) is adapted or vice versa.

The splitting of the input signal into frequency components can take place either by means of a (also non-uniform) filter bank or by means of (short-term) Fourier transformation. Furthermore, this can be split into individual frequency components Signal compared to the un-split signal in time subsampled form are processed.

The inventive combination of a first fixed parametric noise estimator with a second, on the basis of estimates of the first estimator and optionally other criteria time-parametrized Störgeräuschschätzers can be realized a noise estimate, which does not have the disadvantageous characteristics of a fixed parametric noise estimator and does not require the explicit estimation of voice activity , In particular, there is no need to find a compromise between slow adaptation in the presence of a speech signal and fast adaptation when speech is absent. On the contrary, the adaptation of the parameters results in an overall improved noise estimation and thus improved noise reduction, which lessens speech and at the same time interferes with artifacts, e.g. Significantly reduced "musical tones". At the same time, the proposed solution can be implemented efficiently, e.g. by a single time-division multiplexed noise estimator, allowing use in devices with low signal processing capacity, e.g. Hearing aids.

Claims (11)

A method for noise reduction for hearing by Estimating a first value of an input signal (S) with a first estimation algorithm, characterized by Parameterizing a second estimation algorithm with the estimated first value, Estimating a second value of the input signal (S) with the second estimation algorithm and Reducing noise in the input signal (S) based on the estimated second value. The method of claim 1, wherein the first estimation algorithm is based on a minimum tracking method. Method according to Claim 1 or 2, wherein a time change rate of the input signal (S) is estimated by the first estimation algorithm as the first or further value for the parameterization of the second estimation algorithm. Method according to one of the preceding claims, wherein the first estimation algorithm and the second estimation algorithm are structurally the same. Method according to one of claims 1 to 3, wherein the first estimation algorithm includes a recursive smoothing and the second estimation algorithm is not recursive. Method according to one of the preceding claims, wherein the first value is a signal power, an interference power or a signal-to-interference ratio. Method according to one of the preceding claims, wherein in each case a first value is selectively estimated by the first estimation algorithm for a plurality of frequency ranges and these first values are combined to parameterize the second estimation algorithm. Method according to one of the preceding claims, wherein the parameterization of the second estimation algorithm takes place dynamically with a constantly updated first value. Method according to one of the preceding claims, in which the input signal is split into individual frequency components and processed in the time-undersampled form with respect to the un-split signal. Hearing device with a first estimator (NS1) for estimating a first value of an input signal (S) with a first estimation algorithm and a noise reduction device (NR) for reducing a noise in the input signal (S),
marked by - a second estimator (NS2) parameterized with the estimated first value, for estimating a second value of the input signal (S) with a second estimation algorithm, wherein - the noise reduction means (NR) receives the estimated second value from the second estimation means (NS2) for reducing the noise. Hearing apparatus according to claim 10, wherein the first and second estimating means (NS1, NS2) are realized by a single estimating means which is time-divisionally operable alternately as the first and second estimating means.

Images