EP1339256B1

EP1339256B1 - Method for manufacturing acoustical devices and for reducing wind disturbances

Info

Publication number: EP1339256B1
Application number: EP03004661.9A
Authority: EP
Inventors: Hans-Ueli Roeck; Silvia Allegro; Franziska Pfisterer
Original assignee: Sonova AG
Current assignee: Sonova Holding AG
Priority date: 2003-03-03
Filing date: 2003-03-03
Publication date: 2017-12-27
Anticipated expiration: 2023-03-03
Also published as: EP2254349A3; EP2249586A2; EP2249586A3; EP2254351A2; EP1339256A3; EP2254352A3; EP2254350A3; EP2254350A2; DK1339256T3; EP1339256A2; EP2254352A2; EP2254351A3; EP2254349A2

Description

The present invention departs generally from the need of canceling wind disturbances from desired acoustical source reception as of speech or music etc. Wind noise in hearing devices is a severe problem. Wind noise may reach magnitudes of 100 dB SPL (Sound Pressure Level) and even more. Users of hearing devices therefore often switch their device off in windy conditions, because acoustical perception with the hearing device in windy surrounding may become worse than without the hearing device.
Approaches are known to counteract wind noise by mechanical constructional measures, but cannot eliminate wind noise completely, often even not to a completely satisfying degree. It is well-known that wind noise is a low-frequency phenomenon. Depending upon wind speed, direction of the wind with respect to the device, hair length of the individual, mechanical obstructions like hats and other factors, magnitude and spectral content of wind noise vary significantly. With respect to noise, effects and causes we refer to H. Dillon et al., "The sources of wind noise in hearing aids", IHCON 2000, as well as to I. Roe et al., "Wind noise in hearing aids: Causes and effects", submitted to JASA.
Wind signals at sensing ports or acoustical/electrical input converters of hearing devices mounted with a predetermined spacing are far less correlated than are normal acoustical signals to be perceived, as especially speech, music etc.
One reason is that such normal acoustical signals arrive as more or less planar waves, causing at distant acoustical to electrical input converters time delays which are far predominantly caused by the direction of arrival with which such signals impinge upon the converter. As known to the skilled artisan, this time delay is used in beamformer art, whereby a delayed output signal from one converter is subtracted from the output signal of the other converter. There results at the common output of subtraction a signal which has an amplification characteristic with respect to impinging acoustical signals which is dependent on the direction of arrival DOA of such signals with respect to the converters and is commonly known as beamformer characteristics.
The subtraction of well correlated signals as generated by the above mentioned normal signals to be perceived as of speech or music signals normally leads to the known roll-off behavior of such beamformers. The roll-off behavior or characteristic establishes a frequency dependent attenuation of the beam characteristics. It has a pronounced high-pass character, which considerably attenuates low frequencies which are critical especially for speech perception.
Wind noise signals are not subject to the the roll-off behavior of a beamformer because of their lower correlation even at very low frequencies and considered at at least two spaced apart input converters. Whereas normal signals as speech is attenuated by the roll-off towards low frequencies, wind noise is not. Even worse, wind noise has a further adverse effect on signal transfer of normal signals affecting speech recognition. It masks speech-caused signals due to the "upwards-spread-off masking". Upward-spread-off masking is a phenomenon according to which a signal at a predetermined spectral frequency masks signals at higher frequency increasingly with increasing amplitude.
From the US 2002-0 037 088 A1 as well as from the DE 10 045 197 it is known to tackle the problem of wind noise by detecting such noise at two spaced-apart input converters and use in windy situations only the output signal of one of the omnidirectional converters, thereby in fact switching beamforming off. Further, a static high-pass filter is switched on to further attenuate wind noise.
Nevertheless, many hearing devices do not feature two or more acoustical input converters, so that the detection and elimination of wind noise based on two or more converters is not always possible. Further, as was mentioned above, the spectral shape of wind noise varies significantly in time. Thereby, the spectrum range, where wind noise has an energy i.e. below 10⁴ Hz is exactly that range where a hearing device should be effective, because individuals have often impaired hearing abilities in this range. Attenuating wind noise with a static high-pass filter will either filter too little of the wind noise to maintain normal signal perception, or to such an amount that wind noise is well cancelled, but also normal acoustical signals to be perceived. Switching beamforming off as proposed in the above mentioned documents significantly reduces the overall advantages of a hearing device with beamforming abilities also at higher frequencies.
WO 87/00366 A1 describes an improved noise suppression system and WO 97/10586 A1 discloses a method and system for adaptively reducing noise in frames of digitized audio signals that include both speech and background noise.
It is an object of the present invention generically to provide methods and devices which deal with the above mentioned drawbacks. Although it departs from the specific wind noise problems, some of the solutions according to the present invention may also be applied for improving signal-to-noise ratio more generically with respect to normal acoustical signals as of speech or music signals or for improving beamformer control and/or wind detection. Detailed theoretical considerations to the different aspects of the present invention may be found in the paper from F. Pfisterer for achieving their diploma at the Federal Institute of Technology in Zurich. Because this paper is not published yet and discloses more details which might interest the skilled artisan, this paper is enclosed to the present description as Appendix and shall form an integral part of the description.
The present invention resolves the above mentioned object by manufacturing a specifically tailored hearing device. There is proposed a method for manufacturing such a hearing device according to claim 1. Thereby, establishing the operational connections as mentioned needs clearly not to be performed in a time sequence according to the sequence of the wording of claim 1. The operational connections may at least in part be established between units before they are assembled. Further, it must be
emphasized that the output signal of the filter arrangement is just an improved "picture" of the acoustical signals, specific signal processing as for hearing aid devices is performed downstream the filter arrangement.
By this method there is provided a hearing device at which the high-pass characteristic is adapted to the acoustical situation.
In a most preferred embodiment of this method, the step of establishing operational connection of the output of the filter arrangement to the control input of the high-pass filter is performed via a statistics evaluating unit.

Definition

By the term "statistics evaluation unit" we understand a unit at which the behavior of the input signal is continuously monitored during a predetermined amount of time and there is formed over time a statistical criterion of such signal. Generically the output signal of the statistic-forming unit reacts with a time lag on momentarily prevailing characteristics of the input signal and has thus, generalized, a low-pass characteristic. In fact and as example such statistics-forming and evaluating unit may include LMS-type algorithms (Least Means Square) or other algorithms like Recursive Least Square (RLS) or Normalized Least Means Square (NLMS) algorithms.
In a proposed preferred embodiment the statistics-evaluating unit as provided determines the amount of energy of the signal fed to its input and being indicative of the energy at the output of the filter arrangement. Adjusting the high-pass filter characteristic is performed so as to minimize such energy. Thereby preferably one of the algorithms mentioned above is applied. By adjusting the high-pass characteristic, the cut-off frequency or frequencies and/or attenuation slope or slopes and/or low frequency attenuation may be adjustable. In a further embodiment the statistics forming and evaluation unit may estimate speech intelligibility of the output signal of the filter arrangement e.g. by computing the known speech intelligibility index or may estimate speech quality e.g. by computing segmental SNR. The addressed high-pass filter arrangement is realized with a predictor unit. In fact by means of the low-pass filter - with a preceding delay unit - there is established prediction of evolution of the filter input signal. By comparing the output signal of the low-pass filter with the instantaneously prevailing unfiltered signal, principally as occurring at the output of the input converter arrangement, there results a prediction difference between actual signal and predicted signal. As in a most preferred embodiment the low-pass filter is controlled from the output of the comparing unit via statistics evaluation unit, thus with a relatively long reaction time, the low-pass filter may be adjusted to minimize the difference of prediction and actual signal, nevertheless substantially maintaining the spectrum of acoustical normal signals as of speech and music substantially less attenuated. By means of high-pass filter characteristic adjustment the device manufactured becomes optimally adapted to time-varying wind situations.
In a further most preferred embodiment which is especially applied in combination with the above mentioned predictor technique there is provided an analog to digital conversion unit, which is operationally connected at its input side to the output of the input converter arrangement and operationally connected at its output side to the input of the addressed high-pass filter arrangement. Thereby, the said filter arrangement is construed as a digital filter arrangement.
A hearing device, which resolves the above mentioned object is disclosed in claim 5. Further preferred embodiments of such device are disclosed in the claims and the detailed description.
The present invention further resolves the above mentioned object by the method of reducing disturbances, especially wind disturbances, according to claim 9. In a preferred mode generating the third signal in dependency of the second signal, includes performing a statistical evaluation on the second signal, and the third signal is generated in dependency of the result of the statistical evaluation. Thereby, in a still further preferred embodiment the energy of the second signal is evaluated and adjusting of the high-pass characteristic is performed so as to minimize this energy.
Most preferably and especially in the last mentioned realization form, filtering and adjusting is performed digitally.
By the methods and the device according to the present invention, irrespective whether an input acoustical/electrical converter arrangement has one or more than one acoustical/electrical input converters, wind noise is substantially canceled adaptively to the prevailing wind noise situation. Thereby, the signal components to be perceived as resulting from speech or music are substantially less attenuated than wind noise components. Whenever statistic forming and evaluation is performed on basis of a correlation, in a preferred embodiment the statistics forming and evaluation unit has a further input which is operationally connected to the input of the filter arrangement.
Under an additional aspect the present invention deals most generically with further improving signal-to-noise ratio at a hearing device. Thereby, and as will be explained this part of the invention is most suited to reestablish improved signal-to-noise ratio with respect to wind noise after a signal has been processed by high-pass filtering as was explained under the first aspect of the invention.

Definition:

We understand under a "pitch" spectral peaks or peaks of narrow band-width. The fundamental and the spectral harmonics of a signal represent such "pitches".
A pitch-filter is comb-filter with a multitude of narrow pass-bands. It covers for a signal with fundamental and harmonic spectral lines all predominant lines or a predetermined number thereof with pass-bands.
Thereby, the following steps are performed:

providing a pitch filter with adjustable pitch position and with a control input for the pitch position and further with an input and with an output;
providing a pitch detector arrangement with an input and with an output, and
establishing operational connection between the electric output of the input converter arrangement and the input of the pitch filter and between the output of the input converter arrangement and the input of the pitch detector arrangement, and further between the output of the pitch detector arrangement and the control input at the pitch filter, and further establishing the operational connection of the output of the input converter arrangement and the input of the pitch filter with adjustable pitch as well as the operational connection between the output of the input converter arrangement and the pitch detector via the filter arrangement with adjustable high-pass characteristic.

We draw the attention on the WO 01/47335 with respect to pitch filter appliance, which accords with US application No. 09/832 587 .
Generically by means of the pitch detector discrete frequency components in the signals output from the input converter arrangement are detected and their specific frequencies monitored. By controlling pitch position of the pitch filter, i.e. spectral position of its pass-bands, to track the frequencies as monitored, SNR of pitches to noise in the processed signal is improved. Thereby, such pitch signal components are amplified relative to the spectrally intermediate noise.
It has to be emphasized again that establishing the operational connection in the method of manufacturing the hearing device with the pitch filter may be done at least in part well in advance of assembling the units to form the device whenever pitch detection is to be performed by a recursive method, in a preferred embodiment a further input of the pitch detector is operationally connected to the output of the pitch filter.
In one embodiment the hearing device according to the invention comprises

a pitch filter with adjustable pitch position and a control input for said pitch position, further having an input and an output;
a pitch detector unit with an input and with an output, whereby the output of the input converter arrangement is operationally connected to the input of the pitch filter, the output of the input converter arrangement is further operationally connected to the input of the pitch detector unit, and the output of the pitch detector unit is operationally connected to the control input at the pitch filter, whereby the operational connection of the output of the input converter arrangement to the input of the pitch filter and to the input of the pitch detector unit comprises the filter arrangement with adjustable high-pass characteristic.

The method for reducing disturbances according to the invention is performed in one embodiment comprising pitch filtering a first signal dependent from an output signal of an acoustical/electrical input converter arrangement, monitoring the actual pitch frequencies of predominant frequency components within the first signal and adjusting the pitch position of the pitch filtering dependent on the actual pitch frequency positions as monitored, whereby performing pitch filtering is performed on a signal dependent on the second electric signal.
As was already mentioned above, by the technique according to the present invention the signal components to be improved as resulting from speech or music may be attenuated to some extent by high-pass filtering. By the addressed combining SNR with respect to wind noise is further improved. This is realized by first operating the invention with adjustable high-pass filtering upon a signal dependent from the output signal of the input converter arrangement and operating on a signal dependent on the output signal of such high-pass filtering the technique of pitch filtering with controllably adjustable pitch frequency position.
There is further provided improved SNR ratio especially with respect to speech signals.
With respect to spectrum, one characteristic of speech signals is that the fundamental is approximately between 50 Hz and 1 kHz.
Thus, one embodiment of the method for manufacturing a hearing device according to the invention comprises:

providing an adding unit with at least two inputs;
providing a first band pass filter unit with an input and with an output and with a band selected to pass selected harmonics of speech;
a non-linear modulation unit with an input and with an output;
a second band pass filter unit or a low-pass filter unit with an input and with an output and with a pass-band selected on a different harmonics of speech,

from the output of the input converter arrangement to one input of the adding unit without substantial frequency filtering;
from the output of the input converter arrangement to the input of the first band pass filter unit without substantial frequency filtering;
from the output of the first band pass filter unit to the input of the non-linear modulation unit and from the output of the non-linear modulation unit to the input of the second band pass or low-pass filter unit and finally from the output of the second band pass or low-pass filter unit to the second input of the adding unit, whereby the step of establishing operational connection of the output of the input converter arrangement and the one input of the adding unit as well as to the input of the first band-pass unit is performed via the filter arrangement with adjustable high-pass characteristic.

By manufacturing a hearing device as stated the following is realized:

On the output signal of the input converter arrangement speech signals shall be present also and especially with their fundamental components. Due to band-restricted noise as e.g. and especially wind noise, SNR greatly varies considered along the pitches of speech. By selecting at the first band pass filter unit a pass-band according to a harmonics of speech at which a good SNR prevails and subjecting such band filtered signal to a non-linear modulation, all harmonics are regenerated with good SNR. From all the harmonics generated by the non-linear modulation one or more than one band is selected by respective one or more than one second band pass filters or a low-pass filter. The resulting, remaining selected harmonics may first be amplified if desired and are added to the original fundamental and/or harmonics. Thus, in the resulting signal pitches of speech with originally low SNR are improved with respect to that SNR.

Thereby, in a further embodiment an analog to digital conversion unit is provided with an input and with an output, and there is established the operational connection between the output of the input converter arrangement and the one input of the adding unit as well as to the input of the first band pass filter via such analog to digital conversion unit. Thereby, the filter units, the non-linear modulation unit and the adding unit are realized as digital units.
An embodiment of the hearing device according to the invention comprises an acoustical/electrical input converter arrangement with an output, a first band pass filter unit with an input and with an output and with a band selected to pass selected harmonics of speech, a non-linear modulation unit with an input and with an output, a second band-pass filter or low-pass filter unit selected to pass different selected harmonics having an input and an output. There is further provided an adding unit with two inputs and with an output. The output of the input converter arrangement is operationally connected to a first input of the adding unit, substantially without frequency filtering, the output of the input converter arrangement is further operationally connected to the input of the first band pass filter unit, whereby the output of that unit is operationally connected to the input of the non-linear modulation unit. The output of the non-linear modulation unit is operationally connected to the input of the second band pass filter or of the low-pass filter unit, the output of which being operationally connected to the second input of the adding unit. Thereby, the output of the input converter arrangement is connected to the one input of the adding unit as well as to the input of the first band-bass unit via the filter arrangement with the adjustable high-pass characteristic.
Again, preferred embodiment of that device are disclosed in the claims and the specific description.
Further, one embodiment of the method of reducing disturbances according to the present invention, thereby increasing signal-to-noise ratio at a hearing device and especially with respect to speech signals with an acoustical/electrical input converter generating a first electric signal comprises the steps of

band pass filtering a signal dependent on said first signal to generate a band pass filtered signal with harmonic components of speech;
modulating said filtered signal at a non-linear characteristic to generate an output signal with a re-increased number of harmonic components of speech;
band- or low-pass filtering said output signal with said re-increased number of harmonic components to generate a further signal with selected harmonic components and superposing said further signal to a signal dependent on said first electric signal, wherein the step of generating the signal dependent on the first signal comprises filtering the first electric signal with the variable high-pass characteristic.

The invention shall now be described in more details and referring to examples and with the help of figures.
The figures show by examples:

Fig. 1: wind spectra in dependency on wind direction;
Fig. 2: by means of a simplified schematic functional block/signal flow representation a hearing device operating according to the method of reducing disturbances and manufactured by a method, all according to the present invention;
Fig. 3: in a more detailed, but still simplified schematic functional block/signal-flow representation, a preferred embodiment of the invention of fig. 2;
Fig. 4: in a simplified schematic functional block/signal-flow representation an acoustical device which operates a method for improving signal-to-noise ratio and is manufactured by a method;
Fig. 5: in a simplified schematic functional block/signal-flow representation a preferred embodiment of the invention;
Fig. 6: an acoustical device operating a method for increasing signal-to-noise ratio;
Fig. 7: simplified spectra for explaining functioning of the device and method as shown in fig. 6, and
Fig. 8: in a simplified functional block/signal-flow diagram a preferred embodiment of the invention.

In fig. 1 there is shown wind noise spectral characteristic for a wind speed of 10 m/s at an individual head with no hair. Therefrom it might be seen that wind noise spectrum varies significantly as wind direction alters with respect to a device registering such noise. Nevertheless, wind noise spectrum is band-limited.
In fig. 1 there is further schematically introduced the approximate frequency band for human speech fundamental pitch.
In Fig. 2 there is shown, by means of a simplified schematic signal-flow/functional block diagram, an acoustical device, especially a hearing device as manufactured according to the present invention. The device as shown performs the method according to the present invention.
The device comprises, assembled into a schematically shown device casing 1, an input acoustical/electrical converter arrangement 3. Such arrangement 3 may comprise one or more than one specific acoustical/electrical converters as of microphones. It provides for an electric output at A₃, whereat the arrangement 3 generates an electric signal S₃. Possibly via some signal processing, as e.g. pre-filtering and amplifying (not shown), a signal S₃' dependent on S₃ is fed to input E₅ of a high-pass filter arrangement 5. The filter arrangement 5 has a control input C₅ for control signals SC₅ which, applied to C₅, control the high-pass characteristic as shown in block 5 and with respect to its one or more than one corner frequencies f_c, its low-frequency attenuating, one or more than one attenuation slopes. The high-pass filtered signal S₅ output at an output A₅ and is operationally connected, possibly via further signal processing, especially as will be described in context with the second aspect of the present invention, to one or more than one electrical/mechanical output converter arrangements 7 of the device.
With an eye on manufacturing such device all the units as of 3, 5, 9, 7 will be assembled in a casing, whereby they need not be all assembled in the same casing 1, wherein the input converter arrangement 3 is provided. Further, the addressed operational signal connection may be established during or after assembling of the device, some or even all of them may nevertheless be preassembled as by combining units by an integration technique.
A signal S₅" dependent on signal S₅ as output by high-pass filter unit 5, possibly made dependent via additional signal processing as e.g. amplification, is fed from the output A₅ to an input E₉ of a unit 9, which most generically performs upon the signal S₅" a statistical evaluation. The statistic-forming unit 9 performs registering and evaluating selected characteristics of signal S"₅ over time. There results from performing such statistical evaluation that the signal S₉ has a low-pass-type dependency from signal S"₅ input to unit 9. The output signal S₉ at output A₉ is operationally connected, possibly by some intermediate additional signal processing, as e.g. amplification or filtering, to the control input C₅ as a control signal SC₅ and controls the high-pass filter characteristic HP of filter unit 5. As shown in fig. 2, whenever the improved audio signal as of S₅ has to be further processed so as to take individual hearing improvement needs into account, so as customary for hearing aid devices, such processing is performed downstream S₅ at a processor unit PR.
In spite of the fact that functioning of the most generic embodiment as of fig. 2 might be better understood when reading the following explanations to fig. 3 with respect to a preferred form of realization, it is already clear from the embodiment of fig. 2, that, with an eye on fig. 1, the high-pass filter arrangement 5 provides for attenuating wind noise has its corner frequency f_c set and adjusted adjacent the upper end of the wind noise spectra, i.e. somewhere between 1 kHz and 10 kHz. The unit 9 generates the output signal S₉ which does not vary in time on the basis of short-term single signal variation of S"₅, but only with long-term or frequency variations and thereby controls the filter characteristics of filter arrangement 5 to optimize attenuation of such long-term or frequent variations, i.e. signal components as resulting from wind noise. Signal components in S"₅ resulting from normal acoustical signals not to be canceled as from speech or music and appearing in S"₅ with spectra rapidly changing in time will substantially not be canceled by the filter arrangement 5, at least substantially less than steadily or slowly varying or repeatedly occurring signal components as caused by wind noise.
In fig. 3 there is shown a most preferred form of realization of the device and method as disclosed with the help of fig. 2 and accordingly of manufacturing a respectively operated hearing device.
Thereby, signal processing is realized by digital signal processing. Functional blocks and signals, which have already been explained in context with fig. 2 are shown in fig. 3 with the same reference numbers. The output signal S₃ of input converter arrangement 3 is analog/digital converted by an analog/digital conversion unit 11. The filter arrangement 5 as of fig. 2 is realized by a digital filter unit 13. The signal S₃' as input according to fig. 2 to the filter arrangement 5 is now digital and applied to the input E₁₃ of digital HP-filter unit 13. The high-pass - HP - filter arrangement 5 is realized making use of a predictor 15. It comprises a time delay unit 19 and a low-pass digital filter 17, which may be of FIR or IIR type and may be of any particular implementation, e.g. of lattice, direct form, etc. structures.
Signal samples x(n) from input signal S'₃ are input to time delay unit 19, at its input E₁₉. Delayed samples x(n-1) at output A₁₉ of unit 19 are input at input E₁₇ to low-pass filter unit 17, whereat the samples are low-pass filtered to generate at an output A₁₇ an output signal p(n). The units 19 and 17 represent as known to the skilled artisan a predictor and the output signal p(n) is the prediction result.
The prediction result p(n) is compared by subtraction at a subtraction unit 21 with the actual sample x(n) of the actual input signal according to S'₃. Thereby, the output A₁₇ of filter unit 17 is operationally connected to one input of comparing unit 21, the other input thereof being operationally connected to the input E₁₃ of high-pass filter unit 13 without substantial frequency filtering. A matching time delay unit may be introduced in the connection from input E₁₃ to the one input of unit 21 as shown in dashed lines at 22.
At the output A₂₁ of the comparing unit 21 the predictor error signal e(n) is generated, which is indicative for the deviation of the prediction result p(n) from actual signal x(n).
The low-pass filter unit 17 has a control input C₁₇. A control signal applied to that input C₁₇ adjusts the coefficients and/or adaption time constants of the digital filter unit 17. The input C₁₇ of low-pass filter unit 17 represents, with an eye on fig. 2, the control input C₅ of the high-pass filter arrangement 5.
The signal S₁₃ according to the predictor error e(n), is on one hand and as was explained in context with fig. 2 operationally connected to at least one electrical/mechanical output converter (not shown here) of the device.
Further, a signal S₁₃", which depends, possibly via some additional signal processing as e.g. amplification, to signal S₁₃ is input to input E₂₃ of statistics forming and evaluating unit 23. In a most preferred embodiment unit 23 monitors the overall energy of the signal S"₁₃. The control signal C₁₇ to the low-pass filter unit 17 is made dependent from the output signal S₂₃ of unit 23, which is representing the overall energy of the input signal S₁₃". Thereby, in fact in the sense of a negative feedback control loop via control input C₁₇, the adaption time constants and/or the filter coefficients of filter unit 17 are adjusted to minimize the energy of signal S"₁₃ and thus of S₁₃. Thereby, LMS type algorithms or other algorithms like Recursive Least Square (RLS) or Normalized Least Means Square (NLMS) algorithms may be used. In a different embodiment the unit 23 may estimate speech signal intelligibility at signal S₁₃" e.g. by computing from that signal speech an intelligibility index. In a still further embodiment, unit 23 may estimate speech signal quality e.g. by segmental SNR computation.
If unit 23 performs evaluation of statistics based on a correlation, and as shown in dotted line at CR in Fig. 3, the input E₁₃ may be operationally connected to a further input E₂₃₂ of statistics forming and evaluating unit 23.
Although the embodiment of fig. 3, as has been explained, operates in time domain, the same principal may be realized in frequency domain.
As the filter unit 17 is adjusted to minimize the energy of e(n), the predictor 19, 17 will reconstitute the predictable parts of signal x(n) as accurately as possible. Therefore, the prediction error e(n) will only contain non-predictable parts of signal x(n). Because wind noise constitutes substantially predictable components of x(n) and, in opposition, signals to be perceived as especially from speech or music, are non-predictable parts of x(n), the wind noise components are canceled from the output signal S₁₃, finally acting upon the output converter 7, whereas speech or music signals, as non-predictable signals, are passed by S₁₃ to the converter 7.
Experiments have shown that the order of the digital filter 17 may be low, preferably below 5^th order FIR. The resulting filter is thus cheap to implement and still very efficient. Such low-order filter has additionally the advantage of allowing relatively fast adaption times, thus enabling tracking fluctuations of wind noise accurately. Further, it has been found that by the disclosed technique, especially according to fig. 3, wind noise is substantially more attenuated than target signals like speech or music, thereby improving comfort and signal-to-noise ratios.
The skilled artisan being taught the invention as to now may find other adaptive filter structure to realize the principal technique as disclosed.
Under this second aspect of the present invention two additional techniques have been invented, one improving signal-to-noise ratio at an acoustical device, especially hearing device, the other one doing so especially with an eye on speech target signals. Both techniques are most preferably combined in the present invention to further improve low-frequency target signals within a frequency band covered by wind noise spectrum.
Fig. 4 shows, by means of a simplified, schematic functional block/signal-flow diagram an acoustical device, especially a hearing device.
According to fig. 4 an input acoustical/electrical converter arrangement 3, which again may be equipped with one or more than one input acoustical/electrical converters as of microphones, provides at its output A₃ the signal S₃.
A signal D₃ which is dependent from S₃, especially preferred dependent by having been processed by an arrangement as was disclosed in context with figs. 2 and 3, is input to a pitch filter unit 30.
The pitch filter unit 30 is a comb filter as schematically shown within the block of unit 30 with a multitude of pass-bands PB. The filter characteristic of the pitch filter unit 30 is adjustable by a control signal SC₃₀ applied to a control input C₃₀. Thereby, especially the spectral positions as of f₁, f₂... of the pass-bands PB are adjusted. A further signal dependent on the signal S₃, preferably with the same dependency as D₃, F₃₂, is input to an input E₃₂ of a pitch detector unit 32.
Whenever signal F₃₂ has pitch components as schematically shown at the frequencies f_S1..., f_S3 exceeding noise spectrum N the pitch detector unit 32 detects the pitch frequencies f_Sx and generates at its output A₃₂ an output signal G₃₂ which is indicative of spectral pitch position, i.e. of the pitch frequency f_Sx of input signal F₃₂.
The output A₃₂ of pitch detector unit 32 is operationally connected to the control input C₃₀ so as to apply there the control signal SC₃₀ which is indicative of spectral pitch positions within signal F₃₂ and thus S₃.
At the adjustable pitch filter unit 30 the spectral positions of the pass-bands PB are thereby adjusted to coincide with the spectral pitch position f_Sx in signal F₃₂ and thus in signal S₃, so that at the output A₃₀ of the adjustable pitch filter unit 30 a signal S₃₀ is generated, whereat the noise spectrum according to N is substantially attenuated, whereas the pitch components are passed.
If the pitch detector unit 32 operates on the basis of a recursive detection technique, a further input E₃₂₂ of unit 32 is operationally connected to the output A₃₀ of pitch filter unit 30.
This is shown in Fig. 4 by dashed lines at RC.
As not shown in fig. 4 again the output signal S₃₀ is further processed by the device specific signal processor, especially to consider individual needs with respect to hearing improvement as was addressed in context with fig. 2 and is finally operationally connected via such possible signal processing to at least one output electrical/mechanical converter 7.
By the technique under this sub-aspect, signal-to-noise ratio of the device is significantly improved.
Again with an eye on the method for manufacturing such a device, establishing operational connections between the respective units may at least to a certain extent be done before assembling such units to the one or more than one device casings, one of them being schematically shown in fig. 1 at reference No. 1.
The teaching according to this aspect may ideally be combined with the teaching of the present invention. This is schematically shown in fig. 5. Thereby, the output A₃ of the input converter arrangement 3 is operationally connected, again preferably via an analog to digital conversion unit (not shown), to the input E₅ of filter arrangement 5, preferably realized according to fig. 3, the output thereof, A₅, being operationally connected to the adjustable pitch filter system 30/32 as of fig. 4. Thereby, the pitch filter unit 30 in a preferred mode of realization will especially be tailored with pass-bands within the wind noise spectrum as of fig. 1, thereby to reestablish pitches, i.e. frequency components of the tracking signals especially of speech or music signals in that spectral band.
Nevertheless, the technique of applying a controllably adjustable pitch filter may be more generically used to reduce signal-to-noise ratio with respect to tracking signals especially at acoustical devices.
The teaching according to this additional aspect is more specifically directed on improving speech signals.
According to fig. 6 an input acoustical/electrical converter arrangement 3 has an output A₃. A signal H₃ which depends from the signal S₃ output from input converter arrangement 3 is fed to a first input E₄₀₁ of an adding unit 40. At a point P along signal transfer path between S₃ and H₃ a signal I₃ is branched off. The operational connection of the output A₃ to the branching point P is thereby, in a preferred mode, established via the high-pass filtering unit as was explained with the help of figs. 2 and 3 as will be explained later. With respect to frequency content there occurs substantially no frequency filtering in the signal transfer path between branching point P and E₄₀₁, which would be different from such filtering of signal I₃. The signal I₃ is input to an input E₄₂ of a band-pass filter unit 42 with a pass-band PB₄₂. At the output A₄₂ of band-pass unit 42 an output signal I₄₂ is operationally connected to an input E₄₄ of a non-linear modulation unit 44.
At unit 44 the input signal I'₄₂ is modulated at a nonlinear e.g. parabolic characteristic. The modulation result signal I₄₄ at output A₄₄ is operationally connected to input E₄₆ of a second band-pass filter or of a low-pass filter unit 46, without significant frequency filtering.
Unit 46 generates at its output A₄₆ a signal I₄₆. A signal I'₄₆ dependent from the signal I₄₆ without significant frequency filtering is applied to the second input E₄₀₂ of adding unit 40, generating at its output A₄₀ the signal S₄₀. This output signal S₄₀ is (not shown) operationally connected to further signal processing units of the acoustical device, especially the hearing device, which accomplishes device-specific and/or user-specific signal processing.
The functioning of the device or method as shown in fig. 6 and thereby specific selection of the filtering characteristics, especially of units 42 and 46, shall be explained with the help of fig. 7.
In fig. 7(a) there is schematically shown on one hand wind noise spectrum N and on the other hand the fundamental of a speech signal and its harmonics 1, 2, 3, .... It may be seen that whereas fundamental and lower harmonics have bad SNR, higher harmonics have increasingly better SNR.
According to fig. 7(b) the pass-band PB₉₂ of unit 42 is selected to pass high SNR harmonics, resulting in I₄₂ as of fig. 7(c).
This signal is subjected at unit 44 to non-linear modulation. As perfectly known to the skilled artisan by such non-linear modulation, e.g. at a parabolic characteristic, new harmonics are produced as generically shown in fig. 7(d), also considering intermodulation products and folding at the zero-frequency axes.
It has to be noted that these harmonics are spectrally located exactly there where the harmonics and fundamental of the original speech signal according to fig. 7(a) are located.
The signal I₄₄ with good SNR or the signal dependent therefrom is fed to unit 46 with a filter characteristic as shown in fig. 7(e), whereat those harmonics within signal I₄₄ according to fig. 7(d) are canceled or filtered out, which do not accord with original speech harmonics according to fig. 7(a) to be improved as shown in fig. 7(e). At adding unit 40 the signal I'₄₆ with the spectrum according to 7(f) possibly amplified is added to the signal H₃ with a spectrum according to fig. 7(a) resulting in an output signal S₄₀ with speech fundamental and lower harmonics significantly improved with respect to SNR, and as shown in Fig. 7(g).
Thus, the pass-band PB 42 of unit 42 is selected to coincide spectrally with a harmonics of speech with relatively good SNR and the characteristic of filter unit46 is selected so that in the resulting signal harmonics are present, which coincide spectrally with the poor SNR fundamental and lower harmonics of speech to be improved with respect to SNR.
The embodiment as shown in fig. 6 may thereby be implemented digitally by providing down-stream A₃ (not shown) an analog to digital conversion unit and further may be implemented by signal processing in frequency or frequency band domain, thereby adding respective time domain to frequency or frequency band domain conversion units.
As further shown in Fig. 6 a delay unit 43 may be provided between point P and input E₄₀₁ to compensate for time delays between P and E₄₀₂.
With an eye on the method of manufacturing a device according to fig. 6 with a device casing 1, the remaining units are provided and assembled in the same casing or in different casings, the operational connections between the different units being established before, at or after assembling the units in the one or more than one casings.
In a most preferred form the technique as disclosed with figs. 6 and 7 is combined with upstream high-pass filtering of the output signals of the input converter arrangement 3, thereby especially preferred with adjustable high-pass filtering as was explained with the help of the figs. 1 and 2 and which accords to the present invention under its first aspect.
This is schematically shown in fig. 8. The system according to this fig. 8 needs not be additionally described, besides of the fact that the system according to fig. 6 between branching point P and output signal S₄₀ is considered residing in unit 50.
Again and with respect to the methods of manufacturing a device under all aspects of the invention, the operational connections between the various units are established preferably at least to a part before assembling the units in respective single or multiple casings. All aspects of the present invention do not address specific processing of electric signals representing audio signals according to specific device and/or individual needs. By the invention according to the present invention it is achieved - beside of wind recognition per se - that the electric signals at the output of an input acoustical to electrical converter arrangement representing audio signals are improved with respect to their relevancy on signals to be tracked as with respect to signal-to-noise ratio and thereby especially signal-to-wind noise ratio.

Claims

A method for manufacturing an acoustical device, especially a hearing device, comprising the steps of:
• providing in a device casing (1) an acoustical/electrical input converter arrangement (3) with an electrical output (A₃) ;

• providing an audio signal processing unit (PR) for establishing audio signal processing of the device according to individual needs and/or purpose of the device, having an input and an output;

• providing at least one electrical/mechanical output converter (7) with an input (E₇) ;

• providing a filter arrangement (5) with adjustable high-pass characteristic, with a control input (C₅) for said characteristic, an input (E₅) and an output (A₅) ;

• establishing the following operational connections:
between said output (A₃) of said input converter arrangement (3) and said input (E₅) of said filter arrangement (5),

between said output (A₅) of said filter arrangement (5) and said control input (C₅) for said characteristic,

between said output (A₅) of said filter arrangement (5) and said input of said processing unit (PR),

between said output of said processing unit (PR) and said input (E₇) of said at least one output converter (7),

characterized in further comprising:
• realizing said filter arrangement (5) with a predictor unit (15) by operationally connecting said output (A3) of said input converter arrangement (3) with said predictor unit (15) with the following structure:
- a time delay unit (19) with an input (E₁₉) operationally connected to said output (A₃) of said input converter arrangement (3);

- an adjustable low-pass filter unit (17) with an input (E₁₇) operationally connected to an output (A₁₉) of said time delay unit (19) and with an output (A₁₇) operationally connected to one input of a subtraction unit (21);
wherein said output (A₃) of said input converter arrangement (3) is operationally connected substantially unfiltered with respect to frequency to a second input of said subtraction unit (21); and
wherein an output (A₂₁) of said subtraction unit (21) is operationally connected to a control input (C₁₇) of said low-pass filter unit (17) for adjusting a characteristic of said low-pass filter unit (17), said control input (C₁₇) of said low-pass filter unit (17) being said control input (C₅) of said filter arrangement (5) and said output (A₂₁) of said subtraction unit (21) being said output (A₅) of said filter arrangement (5).
The method of claim 1, further comprising the step of establishing said operational connection of said output (A₅) of said filter arrangement (5) and said control input (C₅) for said characteristic via a statistic evaluating unit (9).
The method of claim 2, further comprising the step of providing said statistic evaluating unit (9) determining the amount of energy of a signal (S₅) at said output (A₅) of said filter arrangement (5) and further establishing said adjusting for minimizing said energy.
The method of one of claims 1 to 3, further comprising providing in said casing (1) an analog to digital conversion unit (11) and operationally connecting the input of said analog to digital conversion unit (11) to said output (A₃) of said input converter arrangement (3), operationally connecting the output of said analog to digital converter unit (11) to the input (E₅) of said filter arrangement (5) and providing said filter arrangement (5) as a digital filter arrangement (13).
An acoustical device, especially a hearing device, comprising:
• an acoustical/electrical input converter arrangement (3);

• an audio signal processing unit (PR) for establishing audio signal processing of the device according to individual needs and/or purpose of the device and having an input and an output;

• at least one output electrical/mechanical output converter (7) with an input (E₇);

• a filter arrangement (5) with adjustable high-pass characteristic and with an input (E₅) operationally connected to an output (A₃) of said input converter arrangement (3) and further having a control input (C₅) for adjusting said characteristic, wherein said control input (C₅) for adjusting said characteristic is operationally connected to said output (A₅) of said filter arrangement (5), said output (A₅) of said filter arrangement (5) is connected to said input of said processing unit (PR), the output of which being operationally connected to said input (E₇) of said at least one output converter (7),
characterized in that said filter arrangement (5) comprises a predictor unit (15) with the following structure:
• a time delay unit (19) with an input (E₁₉) operationally connected to said output (A₃) of said input converter arrangement (3);

• an adjustable low-pass filter unit (17) with a control input (C₁₇) being said control input (C₅) for adjusting said characteristic and with an input (E₁₇) operationally connected to said output (A₁₉) of said time delay unit (19) and with an output (A₁₇) operationally connected to one input of a subtraction unit (21);

• a second input of said subtraction unit (21) being operationally connected to said output (A₃) of said input converter arrangement (3) without substantial frequency filtering;

• the output (A₂₁) of said subtraction unit (21) being said output (A₅) of said filter arrangement (5).
The device of claim 5, wherein said output (A₅) of said filter arrangement (5) is operationally connected to said control input (C₅) for adjusting said characteristic via a statistic evaluating unit (9).
The device of claim 6, wherein said statistic evaluating unit (9) operates to determine energy of a signal (S₅) at said output (A₅) of said filter arrangement (5), and wherein said high-pass characteristic is adjusted in dependency of said energy for minimizing said energy.
The device of one of claims 5 to 7, further comprising an analog to digital converter unit (11), an input thereof being operationally connected to the output (A₃) of said input converter (3), the output thereof being operationally connected to said input (E₅) of said filter arrangement (5), said filter arrangement (5) being a digital filter arrangement (13).
A method for reducing disturbances, especially wind disturbances, in an acoustical device, especially a hearing device, with an acoustical/electrical input converter arrangement (3) generating a first electrical signal (S₃), the method comprising the steps of:
• filtering a signal dependent from said first electrical signal (S₃) with a variable high-pass characteristic, thereby generating a second electrical signal (S₅);

• adjusting said variable high-pass characteristic by a third signal (SC₅) dependent on said second electrical signal (S₅),
characterized in said step of performing said filtering comprising the steps of:
• delaying said signal (x(n)) dependent from said first electrical signal (S₃) to provide a delayed signal (x(n-1));

• low-pass filtering the delayed signal (x(n-1)) with an adjustable low-pass characteristic;

• subtracting an output signal (p(n)) dependent on a result of said low-pass filtering from a signal dependent from said first electrical signal (S₃) substantially unfiltered with respect to its frequency content,
and wherein adjusting said variable high-pass characteristic is achieved by controlling said adjustable low-pass characteristic dependent on said subtracting.
The method of claim 9, further comprising the step of generating said third signal (SC₅) in dependency of said second electrical signal (S₅) comprising performing a statistical evaluation on said second electrical signal (S₅) and generating said third signal (SC₅) in dependency of a result of said statistical evaluation.
The method of claim 10, further comprising the step of evaluating by said statistical evaluation energy of said second electrical signal (S₅) and adjusting said variable high-pass characteristic in dependency of said energy so as to minimize said energy.
The method of one of claims 9 to 11, further comprising the step of performing said filtering and adjusting digitally.
The method of claim 1, further comprising the steps of:
• providing a pitch filter (30) with adjustable pitch position, a control input (C₃₀) for said pitch position and with an input and an output (A₃₀);

• providing a pitch detector unit (32) with an input (E₃₂) and with an output (A₃₂);

• establishing operational connections between:
said output (A₅) of said filter arrangement (5) and said input of said pitch filter (30);

said output (A₅) of said filter arrangement (5) and said input (E₃₂) of said pitch detector unit (32);

said output (A₃₂) of said pitch detector unit (32) and said control input (C₃₀) for said pitch position.
The acoustical device of claim 5 comprising:
• a pitch filter (30) with adjustable pitch position and a control input (C₃₀) for said pitch position, further with an input and an output (A₃₀);

• a pitch detector unit (32) with an input (E₃₂) and with an output (A₃₂), the output (A₅) of said filter arrangement (5) being operationally connected to the input of said pitch filter (30), the output (A₅) of said filter arrangement (5) being operationally connected to the input (E₃₂) of said pitch detector unit (32), said output (A₃₂) of said pitch detector unit (32) being operationally connected to said control input (C₃₀) for said pitch position.
The method of claim 9 for additionally improving signal-to-noise ratio in an acoustical device, especially a hearing device, comprising pitch filtering a signal dependent from said second electrical signal (S₅), monitoring the actual pitch of predominant frequency components within said dependent signal, adjusting pitch position of said pitch filtering dependent on said actual pitch position monitored.
The method of claim 1, especially for manufacturing a hearing device, comprising:
• providing an adding unit (40) with at least two inputs (E₄₀₁, E402);

• providing a first band-pass filter unit (42) with an input (E₄₂) and with an output (A₄₂) and with a band selected to pass selected harmonics of speech;

• providing a non-linear modulation unit (44) with an input (E₄₄) and with an output (A₄₄) ;

• providing a second band-pass filter unit (46) or a low-pass filter unit, with an input (E₄₆) and with an output (A₄₆) and with a pass-band selected to pass selected different harmonics of speech,
and establishing the following operational connections:
from the output (A₅) of said filter arrangement (5) to one input (E₄₀₁) of said adding unit (40) without substantial frequency filtering;

from the output (A₅) of said filter arrangement (5) to the input (E₄₂) of said first band-pass filter unit (42) without substantial frequency filtering;

from said output (A₄₂) of said first band-pass filter unit (42) to the input (E₄₄) of said non-linear modulation unit (44);

from the output (A₄₄) of said non-linear modulation unit (44) to the input (E₄₆) of said second band-pass filter unit (46) or low-pass filter unit;

from said output (A₄₆) of said second band-pass filter unit (46) or said low-pass filter unit to the second input (E₄₀₂) of said adding unit (40).
The acoustical device of claim 5, especially a hearing device, comprising an acoustical/electrical input converter arrangement (3) with an output (A₃), a first band-pass filter unit (42) with an input (E₄₂) and with an output (A₄₂), and with a band selected to pass selected harmonics of speech, a non-linear modulation unit (44) with an input (E₄₄) and with an output (A₄₄), a second band-pass filter unit (46) or low-pass filter unit selected to pass different selected harmonics of speech, with an input (E₄₆) and with an output (A₄₆), an adding unit (40) with two inputs (E₄₀₁, E₄₀₂) and with an output (A₄₀), wherein said output (A₅) of said filter arrangement (5) is operationally connected to a first input (E₄₀₁) of said adding unit (40) substantially without frequency filtering, said output (A₅) of said filter arrangement (5) is operationally connected to the input (E₄₂) of said first band-pass filter unit (42), the output (A₄₂) of which being operationally connected to the input (E₄₄) of said non-linear modulation unit (44), the output (A₄₄) of which being operationally connected to said input (E₄₆) of said second band-pass filter unit (46) or of said low-pass filter unit, the output (A₄₆) thereof being operationally connected to the second input (E₄₀₂) of said adding unit (40), and wherein said output (A₃) of said input converter arrangement (3) is operationally connected to said one input (E₄₀₁) of said adding unit (40) as well as to said input (E₄₂) of said first band-pass unit (42) via said filter arrangement (5).
The method of claim 9 for additionally increasing signal-to-noise ratio at the acoustical device, especially the hearing device, with respect to speech signals, further comprising the steps of:
• band-pass filtering a signal dependent on said second signal (S₅) to generate a band-pass filtered signal (I₄₂) with harmonic components of speech;

• modulating said band-pass filtered signal (I₄₂) at a non-linear characteristic to generate an output signal (I₄₄) with a re-increased number of harmonic components of speech;

• band- or low-pass filtering said output signal (I₄₄) with said increased number of harmonic components, to generate a further signal (I₄₆) with selected harmonic components and superposing said further signal (I₄₆) to a signal (H₃) dependent on said first electrical signal (S₃), wherein said step of generating said signal (H₃) dependent on said first electrical signal (S₃) comprises filtering said first electrical signal (S₃) with said variable high-pass characteristic.