EP3796676B1 - Method for operating a hearing aid and hearing aid - Google Patents

Description

The invention relates to a method for operating a hearing aid and to a hearing aid.

A hearing aid is generally used to output sound to a hearing aid user. To do this, a hearing aid has a microphone with which sound from the environment, i.e. ambient sound, is recorded. This generates an electrical input signal, which is fed to a signal processor for further processing. This then generates an electrical output signal, which is output to the user as sound via a hearing aid receiver. A hearing aid is typically worn by the user in or on the ear.

A special design of a hearing aid is a hearing aid designed to compensate for a hearing deficit of a hearing-impaired user. In such a hearing aid, the input signal is modified in the signal processing based on an individual audiogram of the user and is typically amplified in order to compensate for the hearing deficit.

The behavior of the hearing aid is usually characterized by one or more operating parameters, which can be adjusted depending on the situation in order to ensure the best possible hearing experience in different environmental situations. In order to adjust the hearing aid depending on the situation, it is necessary to characterize or classify the environmental situation. An important parameter for this is the signal-to-noise ratio of the environment, i.e. the ratio of useful signal to interference signal. A useful signal is a signal that is of interest to the user and is therefore output to them as clearly as possible. should be suppressed, for example the voice of a speaker with whom the user is talking. An interference signal, on the other hand, is a signal that should be suppressed because it covers the useful signal and thus negatively affects its intelligibility. Examples of interference signals are so-called "babble noise", background noise, other speakers with whom the user is not talking, and environmental or machine noises.

The signal-to-noise ratio is not easily accessible because the levels of the useful component and the noise component must be determined separately in order to then determine their ratio. However, since useful signals and noise signals are present at the same time, they overlap and are recorded together by the microphone. The input signal therefore usually contains both a useful component and a noise component. Separating these two components for the purpose of calculating the signal-to-noise ratio is not easily possible. An approximate calculation using other quantities that are more easily accessible can be highly error-prone.

In the WO 2015/189261 A1 describes a noise reduction system comprising: a signal input for receiving a digital audio input signal comprising a target signal and a noise signal, a first signal-to-noise ratio estimator configured to determine respective first signal-to-noise ratio estimates of a plurality of subband signals based on respective subband noise estimate signals and respective subband signals, a second signal-to-noise ratio estimator configured to filter the plurality of first signal-to-noise ratio estimates of the plurality of subband signals with respective time-varying low-pass filters to obtain respective second signal-to-noise ratio estimates of the plurality of subband signals.

Against this background, it is an object of the invention to provide an improved method for operating a hearing aid and a corresponding hearing aid. In particular, the determination of the signal-to-noise ratio in the environment should be improved. In particular, the best possible estimate of the Signal-to-noise ratio. The estimation should be carried out in particular without an explicit separation of useful part and noise part. The object is achieved according to the invention by a method with the features according to claim 1 and by a hearing aid with the features according to claim 13. Advantageous embodiments, further developments and variants are the subject of the subclaims. The statements in connection with the method also apply mutatis mutandis to the hearing aid and vice versa. If method steps are described below, advantageous embodiments for the hearing aid result in particular from the fact that it is designed to carry out one or more of these method steps.

The method is used to operate a hearing aid and is therefore an operating method. During the method, the hearing aid is worn in particular by a user in or on the ear and used to output ambient sound. The hearing aid has a microphone by means of which ambient sound is recorded and converted into an input signal. The microphone is preferably an omnidirectional microphone, i.e. not a directional microphone, and thus in particular has no preferred direction for recording sound. Analogously, the input signal is preferably an omnidirectional signal. The ambient sound is an acoustic signal. The input signal is an electrical signal. The input signal has a useful component and an interference component. The useful component is a signal that is of interest to the user and should therefore be output to them as clearly as possible. The interference component, on the other hand, is a signal that should be suppressed because it covers the useful component and thus negatively influences its intelligibility. The hearing aid also preferably has a signal processing system to which the input signal is fed for further processing. The signal processing then generates an electrical output signal, which is output to the user as sound via a hearing aid receiver.

The method determines the stationarity of the input signal. For this purpose, the hearing aid and in particular its signal processing expediently have a stationarity detector to which the input signal is fed and which outputs the stationarity. Stationarity is generally understood to be a measure of the variability of a signal over time. A signal that changes little over time has a higher stationarity than a signal that changes more significantly in comparison. The stationarity of a signal in general is measured, for example, by measuring the temporal change of a frequency spectrum of the signal and then deriving a value for the stationarity from this. The less and the slower the frequency spectrum changes, the higher the stationarity. Alternatively or additionally, the signal, specifically its frequency spectrum, is examined for one or more predetermined characteristics and the stationarity is determined depending on the presence or severity of these characteristics.

In the method, a signal-to-noise ratio of the input signal is determined depending on a scaling factor, preferably continuously. The signal-to-noise ratio is also referred to as SNR for short. The signal-to-noise ratio is a measure of the relative proportions of the useful component and the interference component in the entire input signal and thus also in the ambient sound. The scaling factor is determined depending on stationarity, namely using a function which specifies the scaling factor depending on the stationarity of the input signal. The function is stored, for example, in a memory of the hearing aid, specifically the signal processing. The function preferably has a value range of 0 to 1 for the scaling factor, particularly preferably from 0.5 to 1. In other words: the function preferably returns a value in the range of 0 to 1, particularly preferably from 0.5 to 1. Other value ranges are also possible and suitable in principle.

The signal-to-noise ratio is used when operating the hearing aid to adjust it to the situation and thus to make it as optimal as possible. In other words: an operating parameter of the hearing aid is set depending on the signal-to-noise ratio. The signal-to-noise ratio is preferably smoothed before use, e.g. by means of a temporal, particularly rolling, averaging.

An essential aspect of the invention is the stationarity-dependent scaling factor, which is used to determine the signal-to-noise ratio depending on the stationarity of the input signal. In this way, the determination of the signal-to-noise ratio is significantly more precise and an improved setting of the hearing aid is achieved.

The invention is based on the assumption that the input signal contains both a useful component and a noise component and that these two components are initially not available separately for calculating the signal-to-noise ratio. Therefore, the signal-to-noise ratio is determined, or more precisely estimated, based on the input signal. The signal-to-noise ratio determined using the method does not necessarily correspond to the The signal-to-noise ratio determined does not necessarily correspond to the actual signal-to-noise ratio, but represents an estimate. In other words, the signal-to-noise ratio is calculated approximately, especially without precise knowledge of the useful component and the interference component.

Environments with a high level of interference are typically loud, i.e. the corresponding input signal has a high level. At such a high level, the interference component is often, but not necessarily, large relative to the useful component, so that the signal-to-noise ratio is low. As a first approximation, a simple level measurement of the input signal can therefore provide a rough estimate of the likely signal-to-noise ratio. However, this approach is problematic because situations are also possible in which the interference component is low, but the useful component itself is very loud in comparison. In this case, the signal-to-noise ratio is high, but so is the level, so that the assessment based on the simple level measurement is correspondingly incorrect.

The above problem is described below using a specific application: in a practical design, the directionality of the hearing aid is set depending on the signal-to-noise ratio. Directionality generally refers to focusing the hearing aid on a specific listening direction while attenuating or masking out other directions. For example, a beamformer is used for this, which has a directional lobe with an adjustable width. The width of the directional lobe is now set depending on the signal-to-noise ratio. The lower the signal-to-noise ratio, the smaller the width is set, so that only signals that come from a specific direction and are predominantly useful signals are output to the user. This blocks out interference signals from other directions. If a single speaker speaks very loudly in an otherwise quiet environment, a small width and thus a high directionality is set due to the high level, although this is not actually necessary. This means that signals outside the directional lobe are lost, although this is beneficial for an overall more natural listening experience. without impairing the comprehensibility of the useful part too much.

In this case, the signal-to-noise ratio is estimated and the stationarity of the input signal is also taken into account, so that the estimate of the signal-to-noise ratio is improved overall. The stationarity gives the estimate an additional dimension, so to speak, which enables differentiation and classification of the environmental situation. Applied specifically to the use case described above as an example, this means: If the noise component is low, but the useful component is very loud, the stationarity of the input signal is low overall, whereas in the case of a loud noise component, the stationarity is high in comparison. Despite a similar level, situations with very different actual signal-to-noise ratios can then be reliably distinguished and the environment is correctly classified. The estimated signal-to-noise ratio is adjusted accordingly using the scaling factor and then corresponds more closely to the actual signal-to-noise ratio. Aside from the explicitly mentioned use case, any setting of the hearing aid that is carried out depending on the signal-to-noise ratio is therefore significantly improved.

How the signal-to-noise ratio is calculated in concrete terms is not essential for the basic concept, but rather it is only important that stationarity is taken into account. With regard to the calculation of the signal-to-noise ratio, a design is used in which an input level of the input signal is measured and in which an estimated noise component of the input signal is determined. The estimated noise component is multiplied by the scaling factor to produce a scaled, estimated noise component. The signal-to-noise ratio is then calculated by forming a difference between the input level and the scaled, estimated noise component and by calculating the signal-to-noise ratio as the ratio of the difference to the scaled, estimated noise component. This procedure is expressed by the following formula: $$\begin{array}{cc}\mathrm{SNR}& =\left(\mathrm{E-sc*N\_est}\right)/\mathrm{sc*N\_est}\\ & =\left(\mathrm{S+N-sc*N\_est}\right)/\mathrm{sc*N\_est}\end{array}$$

E = S+N is the input level, which is composed of the useful component S (signal) and the noise component N (noise). The scaling factor is designated sc, the estimated noise component N_est. The scaled, estimated noise component therefore corresponds to sc*N_est.

Both the input level and the estimated noise component are derived directly from the input signal, in particular without knowledge of the useful component and the noise component taken separately, i.e. there is no separation of the noise component and the useful component.

In principle, it is possible to determine the signal-to-noise ratio by calculating the ratio of input level to estimated noise component, so that the input level is used as an approximation for the actual useful component and the estimated noise component is used as an approximation for the actual noise component: $$\mathrm{SNR}=\mathrm{E}/\mathrm{Nest}=\left(\mathrm{S}+\mathrm{N}\right)/\mathrm{Nest}$$

For a very small useful component compared to the noise component, i.e. for S ≪ N, and assuming that the noise component is approximately equal to the estimated noise component, i.e. N ≈ N_est, this formula only provides positive values for the signal-to-noise ratio, measured in dB. In other words, cases with a negative signal-to-noise ratio (in dB) cannot be represented.

The representation of a negative signal-to-noise ratio is, however, possible in an advantageous embodiment in which the estimated noise component is first subtracted from the input signal: $$\mathrm{SNR}=\left(\mathrm{S}+\mathrm{N}-\mathrm{Nest}\right)/\mathrm{Nest}$$

In addition, the stationarity-dependent scaling factor is conveniently applied to adjust the proportion and influence of the estimated disturbance component, resulting in the formula already mentioned above: $$\mathrm{SNR}=\left(\mathrm{S}+\mathrm{N}-\mathrm{sc}*\mathrm{Nest}\right)/\mathrm{sc}*\mathrm{Nest}$$

In another suitable variant, the scaling factor in the denominator is omitted and the estimated useful component in the numerator is simply divided by the estimated noise component. Using the scaling factor in the denominator does lead to an additional offset, but this is small. On the other hand, using the scaling factor in the denominator advantageously simplifies handling and implementation of the calculation, since only two variables are then required to calculate the signal-to-noise ratio, namely the input level and the scaled, estimated noise component.

The scaling factor allows the signal-to-noise ratio to be determined more precisely and estimated with less error. If the useful component is larger than the noise component, it is advisable to make no correction or only a small correction using the scaling factor. However, the smaller the useful component is compared to the noise component, the higher the stationarity of the input signal overall and the more it is dominated by the noise component. Here, a stronger compensation is required in order to represent a negative signal-to-noise ratio if necessary. Accordingly, a larger scaling factor is applied with greater stationarity, so that the estimate of the useful component, which is expressed by the numerator (S + N - sc*N_est), is corrected downwards more.

The hearing aid preferably has a first level meter, with which the input level is determined, and a second level meter, in particular a separate one, with which the estimated noise component is determined. The input signal is therefore fed to two different level meters. The level meters are in particular parts of the signal processing. The input level is measured with one level meter, the noise component in the input signal is estimated with the other level meter, with the second level meter being set in such a way that it primarily measures the level of the noise component, i.e. responds less strongly to the useful component than to the noise component. The two level meters are therefore configured differently in order to carry out different level measurements on the same signal, namely the input signal. Overall, only two level meters and one stationarity detector are required to determine the signal-to-noise ratio, each of which is fed with only the input signal. In a practical and particularly simple embodiment, the signal-to-noise ratio is then determined using only two level measurements and one stationarity measurement on the input signal. In an advantageous further development, one or more further measurements are added.

In a suitable embodiment, the estimated noise component is determined using a level meter that is operated with two asymmetrical time constants. This level meter is in particular the previously mentioned second level meter for determining the estimated noise component. By using such an asymmetrical level meter, the level measurement on the input signal is distorted and focused on the noise component.

A particularly advantageous design is one in which the level meter, i.e. in particular the second level meter, is operated with a settling time (attack) that is longer than the release time of the level meter. Such a level meter with a slow settling time and a fast release time is also referred to as a "minimum tracker". The settling time and the release time are each a time constant of the level meter. Due to the settling time being longer than the release time, a corresponding inertia is realized when the level meter responds, which means that the useful component, which is assumed to be less stationary or even non-stationary compared to the interference component, contributes less to the level measurement than the interference component, which is assumed to be stationary compared to the useful component.

The function which specifies the scaling factor depending on the stationarity of the input signal is preferably designed in such a way that a larger scaling factor is determined with greater stationarity of the input signal. In other words: for greater stationarity, the function returns a larger scaling factor. This is based on the consideration that the useful part is rather non-stationary compared to the noise component and vice versa, that the noise component is rather stationary compared to the useful component. A greater stationarity therefore indicates a poorer, ie lower, signal-to-noise ratio. With greater stationarity of the input signal, the proportion of the useful component in the input signal is therefore smaller, so that a larger correction is required, which is then implemented by the larger scaling factor. In a particularly simple embodiment, the function is linear or alternatively linear in sections and otherwise constant.

The function is stored, for example, as a calculation rule or as a table in a memory of the hearing aid, especially the signal processing.

In a practical embodiment, the function is specified by means of a calibration measurement. During the calibration measurement, an actual signal-to-noise ratio is determined for various ratios of a useful component and a noise component and this is compared with the calculated signal-to-noise ratio. The signal-to-noise ratio is preferably calculated using the formula mentioned above, so that the scaling factor then remains as a variable and is determined, in particular according to the following or a similar formula: $$\mathrm{sc}=\left(\mathrm{S}+\mathrm{N}\right)/\left(\mathrm{Nest}*\left(\left(\mathrm{S/<figure-callout\; id="1"\; label="N\; +"\; filenames="imgf0001.png,imgf0003.png"\; state="\{\{state\}\}">N</figure-callout>},\right)+1\right)\right)$$

A known noise component is mixed with a known useful component to obtain an input signal whose actual noise component and actual useful component are thus known. The actual signal-to-noise ratio is then determined according to SNR = S/N and compared with the result of the estimation of the signal-to-noise ratio, and the scaling factor is determined from this. Logically, the estimated noise component is also determined, in particular as provided for in the method. This is now repeated for several different signal-to-noise ratios. The stationarity of the input signal is also determined for each signal-to-noise ratio, so that overall the scaling factor is represented as a function of stationarity.

However, the noise component itself does not necessarily have to be stationary, but can also be non-stationary and, like the input signal as a whole, has a fundamentally variable stationarity. An example of a noise component with low stationarity is so-called "babble noise". An example of a noise component with high stationarity is so-called long-term average speech spectrum, or LTASS for short. Depending on the stationarity of the noise component, the procedural estimate of the signal-to-noise ratio may produce different results, although the actual signal-to-noise ratio SNR = S/N is actually the same. The results typically deviate more, the lower the actual signal-to-noise ratio is. This results in the overall problem that, particularly with an input signal in which the useful component is small compared to the noise component and in which the noise component has low stationarity, the useful component is overestimated and the procedural estimate of the signal-to-noise ratio deviates from the actual signal-to-noise ratio, in particular it is too high, i.e. the signal-to-noise ratio is overestimated. This is particularly related to the estimation of the estimated noise component, since when determining it with the level meter described above, stationary components are primarily taken into account. A strongly non-stationary noise component is therefore only recorded incompletely or not at all, so that the noise component is increasingly underestimated as its stationarity decreases. This problem is solved in an advantageous embodiment by adjusting the function for the scaling factor depending on the stationarity of the noise component. The scaling factor is therefore determined on the one hand depending on a first stationarity, namely the stationarity of the input signal as a whole, and on the other hand also depending on a second stationarity, namely the stationarity of the noise component. The stationarity of the noise component itself is not necessarily measured specifically, but is expediently determined indirectly by determining an input dynamic of the input signal and then assuming that the stationarity of the noise component is greater with lower input dynamics. In other words, the stationarity of the noise component is suitably determined by assuming that below a threshold value for an input dynamic of the input signal, a stationary noise source is present and the noise component is thus stationary, i.e. has a certain stationarity.

Advantageously, the function is adapted depending on the stationarity of the disturbance component in such a way that the function returns a larger scaling factor for a lower stationarity of the disturbance component, i.e. the scaling factor is corrected upwards so that the scaled, estimated disturbance component is larger with decreasing stationarity and the underestimation of the disturbance component is corrected.

The stationarity of the noise component is determined in a suitable embodiment by analyzing the temporal dynamics of the input signal (i.e. the input dynamics), namely by determining a maximum level and a minimum level of the input signal and comparing them with one another. For this purpose, a third and a fourth level meter are expediently used, to which the input signal is fed. The third level meter measures the maximum level, while the fourth level meter measures the minimum level, or vice versa. For this purpose, the two level meters are expediently operated with asymmetrical time constants on the one hand and with time constants that are opposite to one another on the other. This means that the level meter that measures the maximum level is operated with a short settling time and a long decay time, and the level meter that measures the minimum level is operated conversely with a long settling time and a short decay time.

For example, the difference between or the ratio of the maximum level and the minimum level is then determined. The maximum level and the minimum level are preferably determined continuously within a concurrent time interval. In this way, the stationarity of the noise component is advantageously determined based on the input signal without having to know the noise component itself. This takes advantage of the fact that, especially with a low actual signal-to-noise ratio, a higher stationarity of the noise component leads to a lower difference between the maximum level and the minimum level. In other words: the smaller the difference, the higher the stationarity for a given signal-to-noise ratio. The statements apply analogously when using the ratio of maximum level and minimum level. The ratio or the difference are used in one embodiment directly as a measure of the stationarity of the disturbance component.

It is particularly useful to use several different functions which are optimized for disturbances with different stationarities.

In an expedient embodiment, the function for the scaling factor is adapted depending on the stationarity of the interference component by selecting the function for the scaling factor from at least two basic functions depending on the stationarity of the interference component. Depending on the stationarity, one of several basic functions is therefore selected in order to obtain a scaling factor that is optimal depending on the environmental situation. In a particularly simple embodiment, two basic functions are present, a first basic function for stationary or predominantly stationary interference components and a second basic function for non-stationary or predominantly non-stationary interference components. As part of the method, the stationarity of the interference component is first determined, in particular from the input signal as already described. Depending on the stationarity, one of the basic functions is then selected and used as a function to determine the scaling factor. Preferably, the basic function for stationary or predominantly stationary interference components is switched over or faded over as soon as an input dynamic of the input signal falls below a predetermined threshold value, i.e. is sufficiently low.

As an alternative to the above-mentioned discrete selection from several basis functions, in an equally advantageous embodiment the function for the scaling factor is adapted depending on the stationarity of the disturbance component by mixing the function from several basis functions and depending on the stationarity of the disturbance component. In a suitable embodiment, two basis functions are available for this purpose and the function is determined by mixing the two basis functions with one another in a mixing ratio that depends on the stationarity of the disturbance component. This enables a particularly smooth transition when using different basis functions. The basic functions are conveniently designed as previously described.

To mix the basic functions, the hearing aid, especially its signal processing, has a mixer in a suitable design to which the scaling factors from several basic functions are fed. The mixer then mixes these scaling factors in an appropriate mixing ratio depending on the stationarity and then outputs a scaling factor itself, which is finally multiplied by the estimated noise component in order to determine the scaled, estimated noise component.

The calibration measurement described above is conveniently applied analogously to determine various basis functions. The calibration measurement is then carried out not only for different signal-to-noise ratios, but several times for different signal-to-noise ratios, using a noise component with a different stationarity in each case. In a particularly simple embodiment, the calibration measurement is carried out twice, once with a noise component with low stationarity and once with a noise component with high stationarity, so that the calibration measurement provides two corresponding basis functions.

In a suitable embodiment, the hearing aid has several frequency channels, so that the input signal is divided between these several frequency channels. The frequency channels can then be individually modified by the signal processing. The frequency channels are in particular combined again for output. A filter bank is used, for example, to divide them between the various frequency channels. In total, the hearing aid has in particular at least 2, preferably at least 3 frequency channels and preferably 8 to 128 frequency channels. A design with 48 frequency channels is suitable, for example.

The input signal extends over a certain frequency range, in particular the audible frequency range from 20 Hz to 20 kHz or a subrange of it, preferably from 100 Hz to 12 kHz. The signal-to-noise ratio is now determined either over the entire frequency range of the input signal or only over a partial range.

In a particularly useful embodiment, the hearing aid has several frequency channels as described and the signal-to-noise ratio is calculated for each frequency channel of a subset of the frequency channels as described above, so that several signal-to-noise ratios result, from which an average is then formed, which is an averaged signal-to-noise ratio, which is also referred to as a global signal-to-noise ratio. For each of the subsets of frequency channels, a separate, local signal-to-noise ratio is determined separately, so to speak. When estimating the signal-to-noise ratio, not all frequency channels are explicitly taken into account, but some frequency channels are omitted by only taking into account a subset of the frequency channels. This advantageously makes it possible to limit the estimation of the signal-to-noise ratio to the more relevant frequency channels and thus further optimize the operation of the hearing aid. The average is formed in particular by means of an averaging unit of the hearing aid, specifically the signal processing. The number of frequency channels preferably covers a single, contiguous frequency range, but this is not mandatory. A design in which several averaged signal-to-noise ratios are determined, namely for different frequency ranges, is also suitable.

The determination of the signal-to-noise ratio does not necessarily have to be carried out completely for each of the frequency channels separately. Rather, it is sufficient that individual calculations, determinations, determinations or measurements are carried out frequency-dependently, i.e. for individual frequency channels, with other calculations, determinations, determinations or measurements then being carried out globally, i.e. not frequency-dependently. For example, the input level is determined frequency-dependently and thus separately for each individual frequency channel, but the estimated interference component is determined globally based on the summed input level of all frequency channels. In a In another exemplary and suitable variant, the stationarity of the input signal is determined as a function of frequency, averaged and then the scaling factor is determined and the input level and the estimated noise component are determined globally. A design in which the estimated noise component is not determined globally but as a function of frequency is also suitable.

A particularly useful design is one in which the number of frequency channels covers a frequency range of up to 1.5 kHz, i.e. only low frequencies are taken into account when estimating the signal-to-noise ratio. The idea behind this is that the frequency range mentioned is more relevant to the user's perception of volume than other frequency ranges. However, variants in which other frequency ranges are covered alternatively or in addition are also possible and suitable.

As already indicated, an operating parameter of the hearing aid is set depending on the estimated signal-to-noise ratio. In a preferred embodiment, the operating parameter is a parameter of a beamformer, e.g. a directionality or a width of a beamformer's directional lobe, or a parameter of noise reduction, e.g. an attenuation factor or a filter frequency or a filter frequency band of a filter. The improved determination of the signal-to-noise ratio also improves the setting of the operating parameter and the operation of the hearing aid as a whole. For example, the width of the directional lobe of a beamformer is reduced for larger signal-to-noise ratios, i.e. a spatial filter is narrowed in order to achieve a focus by means of which noise components from the environment are suppressed.

The hearing aid is preferably a hearing aid for compensating a hearing deficit of a hearing-impaired user. In such a hearing aid, the input signal is modified in the signal processing based on an individual audiogram of the user by means of a modification unit and in particular amplified in order to compensate for the hearing deficit. The method described is, however, Can also be used advantageously with other hearing aids, e.g. headphones, headsets, telephones, smartphones and the like.

One or more of the described functions or process steps are implemented in the hearing aid and specifically in its signal processing, particularly in terms of programming or circuitry, or a combination thereof. The signal processing is designed to carry out one or more of the described functions or process steps, for example as a microprocessor or as an ASIC, or as a combination thereof.

Fig. 1

ein Hörgerät,

Fig. 2

eine Funktion für einen Skalierungsfaktor,

Fig. 3

ein tatsächliches und ein geschätztes Signal-zu-Rausch-Verhältnisses für einen stationären Störanteil,

Fig. 4

ein tatsächliches und ein geschätztes Signal-zu-Rausch-Verhältnisses für einen nicht-stationären Störanteil,

Fig. 5

eine Variante der Funktion aus Fig. 2,

Fig. 6

ein tatsächliches und ein geschätztes Signal-zu-Rausch-Verhältnisses für einen stationären Störanteil,

Fig. 7

ein tatsächliches und ein geschätztes Signal-zu-Rausch-Verhältnisses für einen nicht-stationären Störanteil,

Fig. 8

eine Variante des Hörgeräts aus Fig. 1.

In the following, embodiments of the invention are explained in more detail with reference to a drawing. In each case, the following schematically shows:

Fig.1

a hearing aid,

Fig.2

a function for a scaling factor,

Fig.3

an actual and an estimated signal-to-noise ratio for a stationary noise component,

Fig.4

an actual and an estimated signal-to-noise ratio for a non-stationary noise component,

Fig.5

a variant of the function Fig.2 ,

Fig.6

an actual and an estimated signal-to-noise ratio for a stationary noise component,

Fig.7

an actual and an estimated signal-to-noise ratio for a non-stationary noise component,

Fig.8

a variant of the hearing aid Fig.1 .

In Fig.1 an embodiment of a hearing aid 2 is shown. A variant of the hearing aid 2 is shown in Fig.8 shown. During a method for operating the hearing aid 2, it is worn by a user (not shown) in or on the ear and used to output ambient sound. The hearing aid 2 has a microphone 4, by means of which ambient sound is recorded and converted into an input signal I. The microphone 4 is an omnidirectional microphone, so that the input signal I is an omnidirectional signal. The Input signal I has a useful component S (signal) and a noise component N (noise). In the examples shown, hearing aid 2 has a signal processing unit 6, to which the input signal I is fed for further processing. Signal processing unit 6 generates an electrical output signal O, which is output to the user as sound via a receiver 8 of hearing aid 2. In the present case, hearing aid 2 is specifically a hearing aid 2 for compensating for a hearing deficit of a hearing-impaired user. Accordingly, input signal I is modified in signal processing unit 6 based on an individual audiogram of the user by means of a modification unit 10 and in particular amplified in order to compensate for the hearing deficit. However, the concepts described here can also be applied to other hearing aids.

During operation of the hearing aid 2, a stationarity st_I of the input signal I is determined. For this purpose, the hearing aid 2 has a stationarity detector 12, to which the input signal I is fed and which outputs the stationarity st_I. Stationarity is generally understood to be a measure of the variability of a signal over time.

Furthermore, a signal-to-noise ratio SNR of the input signal I is determined depending on a scaling factor sc. The signal-to-noise ratio SNR is a measure of the relative proportions of the useful component S and the noise component N in the entire input signal I and thus also in the ambient sound. The scaling factor sc is determined depending on the stationarity, namely using a function F, which specifies the scaling factor sc depending on the stationarity st_I of the input signal I. Two examples of such a function F are given in the Fig.2 and 5 shown.

In this case, the signal-to-noise ratio SNR is determined, or more precisely estimated, based on the input signal I. The signal-to-noise ratio SNR determined using the method does not necessarily correspond to the actual signal-to-noise ratio SNR_t, but represents an estimate. The Fig. 3, 4 , 6 and 7 show comparisons of the estimated signal-to-noise ratio SNR with the actual signal-to-noise ratio SNR_t, where Determination of the estimated signal-to-noise ratio SNR in the Fig. 3 and 4 the function F from Fig.2 and was used to determine the estimated signal-to-noise ratio SNR in the Fig.6 and 7 the function F from Fig.5 .

The estimated signal-to-noise ratio SNR is used, for example, to set an operating parameter P of the hearing aid 2. The operating parameter P is, for example, a parameter of a beamformer or a parameter of a noise reduction.

How the signal-to-noise ratio SNR is calculated in concrete terms is not essential for the basic concept, but rather it is only important that the stationarity st_l is taken into account. In the embodiments shown here, an input level E of the input signal I is measured and an estimated noise component N_est of the input signal I is determined. The estimated noise component N_est is multiplied by the scaling factor sc, resulting in a scaled, estimated noise component sc*N_est. The signal-to-noise ratio SNR is then calculated by forming a difference between the input level E and the scaled, estimated noise component sc*N_est and by calculating the signal-to-noise ratio SNR as the ratio of the difference to the scaled, estimated noise component sc*N_est. This procedure is expressed by the following formula: $$\begin{array}{cc}\mathrm{SNR}& =\left(\mathrm{E-sc*N\_est}\right)/\mathrm{sc*N\_est}\\ & =\left(\mathrm{S+N-sc*N\_est}\right)/\mathrm{sc*N\_est}\end{array}$$

Both the input level E and the estimated noise component N_est are derived directly from the input signal I, without knowledge of the useful component S and the noise component N. A separation of noise component N and useful component S does not occur.

The numerator in the above formula corresponds to an estimated useful component, the denominator to an estimated noise component, so that an estimated signal-to-noise ratio (SNR) is calculated overall. The formula given also makes it possible to display a negative signal-to-noise ratio (SNR). In a variant not shown, the scaling factor sc in the denominator is omitted in the formula mentioned and the estimated useful component in the numerator is simply divided by the estimated noise component N_est.

In the examples shown, the hearing aid 2 has a first level meter 14, with which the input level E is determined, and a separate, second level meter 16, with which the estimated noise component N_est is determined. The input signal I is therefore fed to two different level meters 14, 16. The noise component N in the input signal E is estimated with the second level meter 16, in that the second level meter 16 is set in such a way that it primarily measures the level of the noise component N, i.e. responds less strongly to the useful component S. The two level meters 14, 16 are therefore configured differently in order to carry out different level measurements on the input signal I. The second level meter 16 is operated here with two asymmetrical time constants, namely with an attack time that is longer than a release time. The second level meter 16 is therefore also referred to as a "minimum tracker".

The functions F in the Fig.2 and 5 , which each specify the scaling factor sc depending on the stationarity st_I, are designed in such a way that with greater stationarity st_I of the input signal E a larger scaling factor sc is determined. In the Fig.2 and 5 the stationarity st_I is plotted horizontally and decreases from left to right. The scaling factor sc is plotted vertically and increases from bottom to top. The functions F reflect the idea that with greater stationarity st_I the proportion of the useful part S in the input signal E is smaller, so that a larger correction is required, which is then implemented by the larger scaling factor sc. The functions F shown here as examples are generally designed in a step-like or ramp-like manner and run approximately linearly on a middle section and otherwise predominantly constant on side sections. The two explicitly shown functions F of the Fig.2 and 5 differ on the one hand in terms of the value range for the scaling factor sc and on the other hand in terms of the position of the middle section, i.e. in which value range for the stationarity st_I the respective function F is approximately linear. In Fig.2 the function F for the scaling factor sc has a value range from 0.51 to 0.8. In Fig.5 the function F for the scaling factor sc has a value range of 0.59 to 0.95 and is overall higher than the function F in Fig.2 .

The function F in Fig.2 was determined by means of a calibration measurement as described in Fig. 3 and 4 The same applies to the function F of the Fig.5 regarding the Fig.6 and 7 . In the respective calibration measurement, the actual signal-to-noise ratio SNR_t is first determined for different ratios of a useful signal S and an interference signal N, which is given in the Fig. 3, 4 , 6 and 7 is plotted horizontally and is given in dB. The actual signal-to-noise ratio SNR_t is then compared with the signal-to-noise ratio SNR calculated according to the above formula and with the respective function F. The calculated signal-to-noise ratio SNR is shown in the Fig. 3, 4 , 6 and 7 each plotted vertically and also given in dB. Shown are several point clouds W, in Fig.3 Specifically 11 pieces, one of which is marked with a circle. Also in the Fig.4 and 7 11 point clouds are visible in each case, Fig.6 In contrast, only 10. In the present case, the point clouds W were in a respective of the Fig. 3, 4 , 6 , 7 the same useful signal S is used and the average level of the interference signal N is gradually increased. A respective point cloud W is obtained by plotting the signal-to-noise ratios SNR, SNR_t for different points in time, whereby the level for the useful signal S fluctuates over time, since the useful signal S is, for example, speech, which varies accordingly over time.

In the Fig.3 and 6 A stationary noise component S was used in each case, namely so-called long-term average speech spectra, or LTASS for short (long-term average speech spectrum). In the Fig.4 and 7 In contrast, a non-stationary noise component S, namely so-called "babble noise", was used. From the figures it is immediately clear that the function F of the Fig.2 is better suited for disturbance components S with high stationarity st_N and that the function F of the Fig.5 is better suited for disturbance components N with low stationarity st_N. How Fig.4 shows, the signal-to-noise ratio SNR for low stationary noise components N is reduced to a low, actual signal-to-noise ratio SNR_t is increasingly overestimated, whereas the estimate for stationary noise component N is very good, as Fig.3 show the Fig.6 and 7 show an inverse result when applying the function F according to Fig.5 . How Fig.7 shows, the estimation of the signal-to-noise ratio SNR for non-stationary noise components N is very good, as Fig.6 shows that the signal-to-noise ratio SNR is underestimated for stationary noise components N.

Depending on the stationarity st_N of the noise component N, the method-based estimate of the signal-to-noise ratio SNR may therefore produce different results, although the underlying actual signal-to-noise ratio SNR_t = S/N is actually the same. It is clear that, particularly for an input signal I in which the useful component S is small compared to the noise component N and in which the noise component N has a low stationarity st_N, the useful component S is overestimated and the method-based estimate of the signal-to-noise ratio SNR is too high. This is also clear with regard to the estimated noise component N_est. When determining this using the level meter 16 described above, stationary components are primarily taken into account, so that a strongly non-stationary noise component N is only incompletely recorded or not recorded at all and the noise component N is increasingly underestimated as the stationarity st_N of the same decreases. This problem is solved here by adjusting the function F for the scaling factor sc depending on the stationarity st_N of the disturbance component N. The adjustment of the function F is such that it returns a larger scaling factor sc for a lower stationarity st_N of the disturbance component N, ie the scaling factor sc is corrected upwards with decreasing stationarity st_N. This is taken into account when comparing the Fig.2 and 5 clearly: the scaling factor is determined according to the function F of the Fig.5 , which is optimized for non-stationary disturbance components N, is chosen to be significantly larger than according to the function F in Fig.2 , which is optimized for stationary noise components N.

With hearing aid 2 in Fig.2 only a single function F is used for the scaling factor sc. In the variant of hearing aid 2 according to Fig.8 against it several different basis functions B are used, which are optimized for noise components N with different stationarity st_N. In the hearing aid 2 of the Fig.8 the function F for the scaling factor sc is then adjusted depending on a stationarity st_N of the disturbance component N by mixing the function F from several basis functions B and depending on the stationarity st_N. In the example shown, two basis functions B are present and the function F is determined by mixing the two basis functions B in a mixing ratio that depends on the stationarity st_N. This ensures a smooth transition when using different basis functions B. For example, the two functions F of the Fig. 2 and 5 each used as a basis function B.

To mix the basic functions B, the hearing aid 2 has Fig.8 a mixer 18, to which the scaling factors sc from several basis functions B are fed. The mixer 18 then mixes these scaling factors sc depending on the stationarity st_N in an appropriate mixing ratio and then outputs a scaling factor sc itself, which is finally multiplied by the estimated disturbance component N_est in order to determine the scaled, estimated disturbance component sc*N_est.

In a variant of the hearing aid 2 (not shown), the function F for the scaling factor sc is adapted depending on the stationarity st_N of the noise component N by selecting the function F for the scaling factor sc from at least two basis functions B, for example the one shown in the Fig.2 and 5 shown functions F.

In one possible embodiment, the hearing aid 2 has several frequency channels, not explicitly shown here, so that the input signal I is divided between these several frequency channels. The signal-to-noise ratio SNR is then determined in the same way for some or all of the other frequency channels. The frequency channels are combined again for output. A filter bank is used, for example, to divide the signals between the different frequency channels. The signal-to-noise ratio SNR is determined, for example, for each frequency channel, a subset of the frequency channels is calculated as described above, resulting in several signal-to-noise ratios SNR, from which an average is then formed in an averaging unit, which is an averaged signal-to-noise ratio SNR, which is also referred to as the global signal-to-noise ratio SNR. For each of the subsets of the frequency channels, a separate, local signal-to-noise ratio SNR is determined separately, so to speak. When estimating the signal-to-noise ratio SNR, not all frequency channels are explicitly taken into account, but some frequency channels are omitted by only taking into account a subset of the frequency channels. For example, the subset of the frequency channels covers a frequency range up to 1.5 kHz, i.e. only low frequencies are taken into account when estimating the signal-to-noise ratio SNR.

The determination of the signal-to-noise ratio SNR does not necessarily have to be carried out completely for each of the frequency channels separately, rather it is sufficient that individual calculations, determinations, determinations or measurements are carried out frequency-dependently, ie for individual frequency channels, with other calculations, determinations, determinations or measurements then being carried out globally, ie not frequency-dependently. For example, in the hearing aid 2 in Fig.2 the stationarity st_I of the input signal I is determined frequency-dependently and only for a subset of the frequency channels, averaged and then the scaling factor sc is determined. The input level E and the estimated noise component N_est are determined globally or frequency-dependently.

The stationarity st_N of the noise component N is determined in the embodiments shown by analyzing the temporal dynamics of the input signal I, namely by determining a maximum level Emax and a minimum level Emin of the input signal I and comparing them with each other. For example, the difference between or the ratio of the maximum level Emax and the minimum level Emin is determined. In this way, the stationarity st_N is determined without having to explicitly know the noise component N. This takes advantage of the fact that, especially with a low actual signal-to-noise ratio SNR_t, a higher stationarity st_N of the noise component N leads to a lower difference between maximum level Emax and minimum level Emin. The function F is then adjusted in such a way that with a larger difference a lower stationarity st_N is assumed and therefore a correspondingly adjusted scaling factor sc is used. In this case, a third and fourth level meter 20 are used, which are fed the input signal I and which determine the maximum level Emax and the minimum level Emin and thus also the stationarity st_N.

List of reference symbols

2

Hearing aid

4

microphone

6

Signal processing

8th

Listener

10

Modification unit

12

Stationarity detector

14

first water level meter

16

second level meter

18

mixer

20

third and fourth level meters

E

Input level

Emax

Maximum level

Emin

Minimum level

F

function

I

Input signal

N

Noise component

Nest

estimated interference

O

Output signal

P

Operating parameters

S

Usable share

sc

Scaling factor

sc*N_est

scaled, estimated noise component

SNR

(estimated) signal-to-noise ratio SNR_actual signal-to-noise ratio

st_I

Stationarity of the input signal

st_N

Stationarity of the disturbance component

W

Point cloud

Claims (13)

1. Method for operating a hearing device (2),

   - wherein the hearing device (2) comprises a microphone (4) by means of which ambient sound is picked up and converted into an input signal (I) comprising a useful component (S) and an interference component (N),

   - wherein a stationarity (st_I) of the input signal (I) is determined, stationarity (st_I) being understood to mean a measure of the variability of a signal over time, a signal that changes little over time having higher stationarity than a signal that changes more sharply in comparison,

   - wherein an input level (E) of the input signal (I) is measured,

   - wherein an estimated interference component (N_est) of the input signal (I) is determined and is multiplied by a scaling factor (sc), resulting in a scaled, estimated interference component (sc*N_est),

   - wherein the scaling factor (sc) is determined according to stationarity, specifically on the basis of a function (F) that indicates the scaling factor (sc) according to the stationarity (st_I) of the input signal (I),

   - wherein a signal-to-noise ratio (SNR) of the input signal (I) is calculated by calculating a difference between the input level (E) and the scaled, estimated interference component (sc*N_est) and by calculating the signal-to-noise ratio (SNR) as a ratio of the difference in relation to the scaled, estimated interference component (sc*N_est) or merely in relation to the estimated interference component (N_est),

   - wherein the signal-to-noise ratio (SNR) is taken as a basis for setting an operating parameter (P) of the hearing device (2).
2. Method according to Claim 1,

   - wherein the hearing device (2) comprises a first level meter (14), which is used to determine the input level (E), and a second level meter (16), which is used to determine the estimated interference component (N_est).
3. Method according to either of Claims 1 and 2,

   - wherein the estimated interference component (N_est) is determined using a level meter (16) that is operated with two asymmetric time constants.
4. Method according to Claim 3,

   - wherein the level meter (16) is operated with an attack that is greater than a release of the level meter (16).
5. Method according to one of Claims 1 to 4,

   - wherein the function (F) is designed such that greater stationarity (st_I) results in a greater scaling factor (sc) being determined.
6. Method according to one of Claims 1 to 5,

   - wherein the function (F) is predefined by means of a calibration measurement in which an actual signal-to-noise ratio (SNR_t) is determined for different ratios of a useful component (S) and an interference component (N) and said actual signal-to-noise ratio is compared with the calculated signal-to-noise ratio (SNR).
7. Method according to one of Claims 1 to 6,

   - wherein the function (F) for the scaling factor (sc) is adapted according to a stationarity (st_N) of the interference signal (N).
8. Method according to Claim 7,

   - wherein the stationarity (st_N) of the interference component (N) is determined by analysing the temporal dynamics of the input signal (I), specifically by determining a maximum level (Emax) and a minimum level (Emin) of the input signal (I) and comparing said levels with one another.
9. Method according to the either of Claims 7 and 8,

   - wherein the function (F) for the scaling factor (sc) is selected from at least two basic functions (B) according to the stationarity (st_N) of the interference component (N).
10. Method according to either of Claims 7 and 8,

    - wherein two basic functions (B) are available and the function (F) is determined by mixing the two basic functions (B) with one another in a mix ratio that is dependent on the stationarity (st_N) of the interference component (N).
11. Method according to one of Claims 1 to 10,

    - wherein the hearing device (2) has multiple frequency channels,

    - wherein the signal-to-noise ratio (SNR) is calculated for each frequency channel in a subset of the frequency channels, resulting in multiple signal-to-noise ratios (SNR) that are then used to calculate a mean value, which is an averaged signal-to-noise ratio (SNR).
12. Method according to one of Claims 1 to 11,

    - wherein the operating parameter (P) is a parameter of a beamformer or a parameter of a noise reduction system.
13. Hearing device (2) designed to carry out a method according to one of Claims 1 to 12.

Images