JPH0834652B2 - Hearing aid system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention is suitable for a hearing-impaired person, particularly a sensorineural hearing-impaired person, and outputs a sound of natural sound quality within the dynamic range of the hearing-impaired person. To an improved hearing aid system.

[Prior Art] Hearing loss is roughly divided into "conducting hearing loss" and "sensory hearing loss"
There are two types. In the case of the former, the "threshold" which is the minimum value of the sound pressure that can be heard and the "discomfort level" which is the maximum value are both higher than those of the hearing-impaired person, and therefore a simple amplifier is sufficient as a hearing aid. However, in the latter case, the discomfort level is almost the same as that of a normal hearing person, and the threshold value often rises. Moreover, since the discomfort level and the change in the threshold value are often different depending on the frequency, the hearing aid is a simple amplifier. In that case, there is a drawback that the output level of the hearing aid does not fall within the dynamic range (threshold to discomfort level) of the hearing impaired person and is too small or too large.

Therefore, recently, in order to be applied particularly to this "sensorineural hearing loss", the threshold value and the discomfort level for each frequency of the hearing impaired person are measured in advance, and the spectrum of the output signal is used to measure the spectrum of the output signal. Hearing aids of the type have been devised that operate to fall within the dynamic range of the threshold and the discomfort level.

One known example of such a type of hearing aid (see Reference 1)
As shown in FIG. 14, frequency analysis of the input signal is performed by a band filter bank, and the obtained level of each band is calculated by comparing the actual value of the threshold and the discomfort level of the deafness with the threshold and the discomfort level of the normal hearing person. Using the loudness mapping function calculated in advance for each band from the measured value, it is converted into a gain for each band, this gain is input to the multiplier for each band to control the amplitude of the signal for each band, and finally these controls The added signals are added over the entire frequency band and output.

In the second known example (see Document 2), as shown in FIG. 15, a short-time Fourier analysis is performed to perform frequency analysis for each block obtained by dividing the input signal into constant intervals, and the obtained constant is obtained. The level for each frequency is converted into a gain for each constant frequency by a loudness mapping function, and this gain is converted to a frequency sampling structured filter [when the gain for a certain discrete frequency (sample frequency) is determined, By using it as a filter coefficient of a filter having a configuration in which a frequency-gain characteristic smoothly interpolated between them is obtained,
By implementing a frequency characteristic filter in which gains given for each constant frequency are smoothly connected, and an input signal is passed through this filter, an output subjected to amplitude control for each frequency is obtained.

Reference 1 JCVentura “Programmable Gain Controller For H
earing Haids, “Procedings of 13th Internatinal Con
gress on Acoustics (Belgrade, 1989) Reference 2 Tetsuya Hayashi, 6 others, “On trial production of master hearing aids by digital method”, Proceedings of Acoustical Society of Japan (March 1989), 2-1-7 [Invention Problem to be Solved] In the first known example, since the signal is divided and processed on the frequency axis by the band filter, the frequency characteristic of the obtained output signal is not smooth but has a stepped shape.
Therefore, the sound quality becomes unnatural.

Further, in the second known example, since the frequency analysis is performed by the short-time Fourier analysis, the frequency analysis result is obtained at constant time intervals. Therefore, the change in the coefficient of the frequency sampling structured filter is not constant at constant time intervals. It occurs continuously, and as a result, the output signal changes discontinuously at regular intervals. Therefore, a click sound may occur in the output.

An object of the present invention is to improve both of the problems of the above-mentioned known technology, that is, the frequency characteristic of the output signal is not smooth and that the output signal changes abruptly with time, to make the sound quality more natural. It is to provide a hearing aid system that can be done.

[Means for Solving the Problems] With the above object, the hearing aid system of the present invention has the structural features described in each of the claims.

[Operation] In the hearing aid system of the present invention according to each of claims 1 to 3, the frequency sampling structure type filter means ensures a continuous smoothness on the frequency axis, and the frequency analysis means ensures a continuous smoothness on the time axis. Further, the means for designating each gain of the frequency sampling structure type filter means based on the loudness mapping function for each frequency has the function of adapting the output sound level to the loudness curve of the hearing-impaired individual for each frequency.

Further, in the hearing aid system according to the fourth aspect, the continuous smoothness on the time axis is provided by the window means acting on the output of the frequency sampling structure type filter means.

Further, in the hearing aid system according to claim 5, the gain for each sampling frequency of each block calculated based on the loudness mapping function is temporally smoothed between the blocks, and the gain is adjusted at each sampling frequency of the frequency sampling structured filter means. By giving it as a gain, smooth operation on the time axis is ensured.

[Examples] FIG. 1 is an overall configuration diagram showing a first example of a hearing aid system of the present invention. The input showing the electric signal waveform of the input sound is converted into a digital signal by an A / D converter that operates at an extremely short sampling period, and the frequency sampling structure type filter 1
After passing through, it is D / A converted and becomes an output signal. This is converted into an output sound and reaches the ear of the wearer of the hearing aid.

As shown in FIG. 2, the frequency sampling structured filter 1 has gains at N discrete frequencies (sample frequencies).
g _1, g ₂ ..., if g _N is specified, the third diagram of such a frequency and smoothly interpolate the street and between these points in the target present frequency - is a filter, such as to realize a gain characteristic.
An example of a formula as a concrete digital filter for realizing this is shown below. Let N be the number of frequency samples. The value of each frequency sample (gain as a complex number at each sample frequency) is (k) (k = 0, 1, ..., N-1). The z-transform transfer function of the desired filter is given by:

However, (Conjugate complex relationship). here, Putting, the signal flow diagram becomes like Fig. 4. In this way, the characteristics of the frequency sampling structure type filter can be controlled both in terms of amplitude and phase, and the constraint is that the characteristics have a conjugate complex relationship as described above, but in reality the phase is zero. And Therefore, in this embodiment, it is assumed that the phase is zero for each gain g ₁ , g ₂ , ... G _N at the sampling frequency.

Now, in the present embodiment, the frequency sampling structure type filter 1
The gains g ₁ , g ₂ , ..., G _N at each sampling frequency are designated by the signal processing route shown in the upper half of FIG. This will be described below.

First, the short-time Fourier analysis means 2 uses the sample values obtained by A / D converting the input signal, and successively short-term intervals (each of which is blocked by the short-time Fourier analysis means 2) partially overlap each other as shown in FIG. Each of these blocks contains a large number of sample values.)
Perform Fourier analysis in. In this short-time Fourier analysis, in each block, each Fourier coefficient (absolute value of each complex Fourier coefficient, that is, each Fourier coefficient) is calculated for each of the N frequency components that are the same as the N sampling frequencies of the frequency sampling structured filter 1. The amplitude of the frequency component, and thus the level, is referred to as a Fourier coefficient in this specification) (FIG. 6). The Fourier coefficient of the i-th frequency component obtained in the n-th block is represented by a _i ⁽ⁿ⁾ (i = 1, 2, ..., N). Fourier coefficients of these sequential blocks
a _i ⁽ⁿ⁾ is, as shown in FIG. 1, temporal smoothing means 3 _i (i =
1,2, ..., N) are input respectively. This temporal smoothing means 3 _i
In, the calculation of Wa _i ^(n-1) + w'a _i ⁽ⁿ⁾ is performed. Here, w and w ′ are weighting functions, and w decreases with time in the overlapping component of the n−1th block and the nth block,
w'changes to increase. FIG. 7 is a diagram illustrating this weighting function (may be expressed as a time window). As a result, as shown in FIG. 8, the Fourier coefficient a _i , in which the temporal change between the blocks is smooth, becomes the temporal smoothing means.
It is continuously output from 3 _{i as} time passes.

The Fourier coefficient a _i of the i-th frequency component obtained from the temporal smoothing means 3 _{i for} each Fourier coefficient is input to the loudness mapping function means 4 _i (i = 1, 2, ..., N). here,
The loudness mapping function will be described below.

Loudness is a quantity that represents a psychological (ie, perceived) loudness of a sound, while sound pressure level is a quantity that represents a physical loudness. There is a difference in the correspondence between the sound pressure level and the loudness between a hearing-impaired person and a hearing-impaired person, and a hearing-impaired person generally has a lower loudness for a sound of the same sound-pressure level than a hearing-impaired person (that is, a sound sounds small). ). In the case of a sensorineural hearing-impaired person, the discomfort level of sound pressure is almost the same as that of a hearing-impaired person, but the threshold value is high. It looks like Figure 9.
From this figure, it is possible to make a graph of the sound pressure level for a hearing-impaired person and the sound pressure level for a hearing-impaired person so that the hearing person has the same loudness as a hearing-impaired person.
It looks like the solid line in Figure 10. This is the loudness map function. Therefore, if the input sound pressure level is converted according to the input / output characteristic shown by the solid line in FIG. 11 which is similar to that in FIG. 10, it is possible to give a hearing-impaired person the same loudness curve as a hearing-impaired person. For that purpose, the gain may be changed according to the input level according to the solid line of FIG. The straight line shown by the broken line in FIGS. 10 and 11 and the horizontal straight line shown by the broken line in FIG. 12 show the case of a normal hearing person as a reference. Loudness curve of the deaf (9th
(Fig.) Is generally different depending on the frequency.Therefore, by performing measurement in advance for a hearing-impaired person who is a hearing aid wearer, the above loudness mapping function for each frequency of the hearing-impaired person, and thus Fig. 11, 12 You can create a function like the one shown.

The loudness mapping function means 4 _i (i = 1, 2, ..., N) in the present embodiment shown in FIG. 1 respectively has N sampling frequencies of the above frequency sampling structured filter 1 (this is the above-mentioned It is also N frequencies derived by short-time Fourier analysis of
From the loudness mapping function of the hearing-impaired person for each of the above, the functional relationship of FIG. 12 for each of these frequencies is calculated. Each loudness mapping function unit 4 _i, in response to the Fourier coefficients a _i input from the smoothing means 3 _i,
The corresponding gain is obtained from the relationship in FIG. 12 (the above Fourier coefficient corresponds to the input level in FIG. 12). Specifically, for example, a correspondence table between the input level and the gain corresponding to the graph of FIG. 12 may be stored in the ROM, and the gain may be obtained using the table. N thus obtained
N gains g ₁ , g ₂ , ..., G _N for the sampling frequencies are given to the frequency sampling structured filter 1.

As described above, according to the embodiment of FIG. 1, it is possible to obtain a hearing aid system that is continuously smooth on both the time axis and the frequency axis and that is well adapted to the loudness curve for each frequency of the hearing-impaired person.

Next, a second embodiment of the hearing aid system of the present invention will be described.
FIG. 13 shows the overall structure. In this embodiment, the first
Instead of performing the short-time Fourier analysis divided into successive blocks as in the embodiment, the frequency analysis is performed continuously in time by the filter bank including the narrow band filters 5 ₁ , 5 ₂ , ..., 5 _N. There is. Accordingly, there is no Fourier coefficient temporal smoothing means 3 ₁ , 3 ₂ , ..., 3 _{N as} in the first embodiment. These bandpass filter 5 _1, 5 _2, ..., pass center frequency of 5 _N is, N pieces of sampling frequency of the sampling structure type filter 1, therefore, the loudness mapping function unit 4 _1, 4 _2, ..., a 4 _N Matched with each frequency. Other configurations and operations are similar to those of the first embodiment. Also in this embodiment, it is possible to obtain a hearing aid system that is smooth on both the time axis and the frequency axis and that is well adapted to the loudness curve for each frequency of the hearing-impaired person.

FIG. 16 shows an outline of the processing of another embodiment of the hearing aid system of the present invention. First, an input signal sampled at a short sampling period is divided into blocks of appropriate length that partially overlap each other, and a short-range Fourier analysis is performed on the signals in this block to obtain the average spectrum in the block (the first embodiment described above). The same thing as the short-time Fourier analysis result in) is obtained. Next, based on the loudness mapping function (a function that indicates the correspondence between the loudness of a hearing-impaired person and a hearing-impaired person) prepared in advance for each frequency, the optimum filter of the frequency sampling structure type filter for this block is selected. The characteristics are determined, and filtering is performed for each block.

The filter coefficient is updated at the boundary between adjacent blocks. At this time, since the amplification factor changes, the output signal becomes discontinuous and may be perceived as a click sound. To prevent this, after filtering, the 16th
As shown in the figure, a time-varying window is provided between adjacent blocks that overlap each other so as to connect them smoothly.

The hearing aid system prototyped based on the embodiment shown in FIG. 16 will be disclosed below. Each block for the short-time Fourier analysis described above has a sample value of 512 points (32m
s), the length of overlap was set to 256 points, but there was no sound quality deterioration due to the click in the voice sample used. The processing side and evaluation experiment of this prototype hearing aid system are shown below.

Processing example As an example of the level fluctuation of the input signal, FIG. 17 shows the change of the filter characteristics and the spectrum of the input and output signals when the input level is changed by 15 dB for the same input signal. From this figure, it can be seen that even if the level of the input signal fluctuates, the frequency characteristic of the filter follows this, and the spectrum of the output signal is mapped within the dynamic range of the hearing impaired person.

Evaluation experiment A clear test was conducted using Japanese monosyllabic (male, 100 syllables). The test subjects were three adult males with normal hearing ears, and in order to simulate hearing loss, masking noise was applied so that the threshold value would be the threshold value of the hearing impaired person in FIG. The input sound level was set to two kinds with relative levels of 0 dB and −15 dB (corresponding to FIGS. 17 (a) and 17 (b)), and presented from one ear using headphones in an anechoic room. The judgment for one syllable is repeated three times. Table 1 shows the results when the hearing aid system was not used and when this hearing aid system was used. From this, it can be seen that, when the relative level of the input voice is 0 dB, the correct answer rate is significantly improved when the hearing aid system is used as compared with the case where the hearing aid system is not used. Even when the input level is -15 dB, it is greatly improved to 0 dB.
Compared with the result of 1., although the correct answer rate was slightly lowered, almost the same result was obtained, and the effect of following the spectrum of the input signal of this system was confirmed.

As still another embodiment of the present invention, the frequency analysis result by the short-time Fourier analysis of each block in the first embodiment is directly applied to the loudness mapping function means 4 without passing through the Fourier coefficient temporal smoothing means 3 _i. given to _i, wherein each of each sample frequencies obtained gain g _1, g _2, as giving ...... g _N to the frequency sampling structure type filter 1 via the means for temporally smoothed between the blocks It is also possible to have different embodiments.

[Advantages of the Invention] The hearing aid system of the present invention can match the loudness curve of each frequency of the individual hearing impaired well and can have smooth operation characteristics without discontinuity on the frequency axis or the time axis. The click sound does not occur and natural sound quality can be provided.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram of a first embodiment of the present invention, FIGS. 2 to 4 are explanatory diagrams of a frequency sampling structure type filter and a signal flow exemplification diagram, and FIGS. Explanatory drawing of the short-time Fourier analysis in an example, FIG. 7 and FIG.
The figure is also an explanatory view of temporal smoothing of Fourier coefficients,
Figures to 12 are explanatory diagrams of the loudness mapping function,
FIG. 13 is an overall configuration diagram of the second embodiment of the present invention, FIGS. 14 and 15 are configuration diagrams of different known examples, and FIG. 16 is a diagram showing a processing configuration of another embodiment of the present invention. 17 Figures (a), (b)
[Fig. 3] is a diagram showing experimental results of the same example.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Tetsuya Hayashi 52 Kawauchi Carpenter Town, Aoba-ku, Sendai City, Miyagi Prefecture 52 Kawauchi Kofuji Fuji Corp.

Claims (5)

[Claims] 1. A frequency obtained by passing a point indicating each gain at the sampling frequency and smoothly interpolating the gain between the sampling frequencies when respective gains are specified for some given sampling frequencies. A frequency sampling structure type filter means exhibiting a gain characteristic, a means for inputting an input signal representing an input sound wave to the frequency sampling structure type filter means, and an output signal passed through the frequency sampling structure type filter means A means for converting into a sound wave, a frequency analysis means for continuously obtaining the level of the input signal for each of the sample frequencies, and a wearer according to the level for each of the sample frequencies obtained by the frequency analysis means. And means for specifying the gains of the frequency sampling structured filter means based on the loudness mapping function for each of the sampling frequencies, respectively. Vessel system. 2. The frequency analysis means includes short-time Fourier analysis means for performing a Fourier analysis for each of the short-time blocks that partially overlap each other with respect to the input signal, and each of the short-time Fourier analysis means. The hearing aid system according to claim 1, further comprising means for temporally smoothing the Fourier coefficient of each sampling frequency in a block between the blocks. 3. The frequency analysis means is a filter bank comprising a plurality of narrow band filters which respectively receive the input signal and pass the respective sampling frequencies. Hearing aid system. 4. A frequency obtained by passing a point indicating each gain at the sampling frequency and smoothly interpolating the gain between the sampling frequencies when the respective gains are specified for some given sampling frequencies. -Frequency sampling structure type filter means exhibiting a gain characteristic, means for inputting an input signal representing an input sound wave to the frequency sampling structure type filter means, and short-time blocks that partially overlap each other for the input signal Based on the loudness mapping function of the wearer for each frequency corresponding to the frequency component according to the level of the frequency component in each block obtained by the short-time Fourier analysis unit Means for designating the respective gains of the frequency sampling structure type filter means, and a signal passed through the frequency sampling structure type filter means. A window means for time varying as smoothly continuously be output between the blocks, the hearing aid system comprising: means for converting an output signal from the window unit to the output waves, in that it consists of. 5. A frequency formed by passing a point indicating each gain at the sampling frequency and smoothly interpolating the gain between the sampling frequencies when respective gains are specified for some given sampling frequencies. A frequency sampling structure type filter means exhibiting a gain characteristic, a means for inputting an input signal representing an input sound wave to the frequency sampling structure type filter means, and an output signal passed through the frequency sampling structure type filter means Means for converting to sound waves, short-time Fourier analysis means for performing Fourier analysis for each short-time block that partially overlaps the input signal, and frequency components in each block obtained by the short-time Fourier analysis means To the frequency sampling structure type filter means based on the loudness mapping function of the wearer for each frequency corresponding to the frequency component according to the level of Means for calculating each gain for each sampling frequency in each block, and the gains calculated by the means are smoothed temporally between the blocks, and then the frequency sampling structure type filter means is provided. A hearing aid system comprising: means for giving.