JP2563719B2 - Audio processing equipment and hearing aids - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice processing device which selectively emphasizes an important frequency band of input voice.

The present invention also relates to a hearing aid equipped with a voice processing device for selectively emphasizing an important frequency band of input voice.

[0003]

2. Description of the Related Art In speech, each phoneme is characterized by several energy bands, called formants, in which energy is concentrated. Humans perform frequency analysis on speech in the cochlea and auditory nerve of the inner ear, and use the distribution of each formant frequency as one clue to identify the phoneme. However, a hearing-impaired person often has a reduced ability to discriminate between sounds having different frequencies (simultaneously existing frequencies) (decrease in frequency resolution), which makes it difficult to perceive formants. Further, even in a normal hearing person, in a noisy environment, the frequency resolution is lowered due to the masking effect of noise, and the clarity of the voice is lowered.

A formant emphasizing device is an example of a voice processing device for coping with such a decrease in frequency resolution. This is intended to improve the frequency resolution by increasing the contrast of the energy distribution of the formant and the other frequency bands.

As a conventional voice processing device, for example, 19
It is shown in the 90th IEICE Spring National Convention A-253. FIG. 9 shows a block diagram of this conventional voice processing apparatus. In FIG. 9, white arrows indicate that a plurality of data are transferred. In FIG. 9, 10 is a filter bank for dividing the input signal into a plurality of bands, 20 is a rectifying / integrating means for rectifying and integrating the band-divided signals for each band to obtain a short-time average amplitude, and 30 is a filter for each band. Convolution integration means for convolving the average amplitude with a function of a specific shape generated by the function generation means 40 on the frequency axis at each sampling time. Reference numeral 50 is a normalizing means for normalizing by dividing the signal after convolution integration in each band for each sampling time by the signal before convolution integration to determine the gain of each band. Reference numeral 60 denotes a multiplying unit that multiplies the output signal of the filter bank 10 and the gain of each band which is the output of the normalizing unit to generate an output signal of each band. Reference numeral 70 denotes a summing means for calculating the sum of the output signals of each band and outputting the sum.

The operation of the conventional speech processing apparatus configured as above will be described. In FIG. 9, the function generating means 40
Then, for example, the side suppression function that combines the error functions

[0007]

[Equation 1]

When the above is generated, the spectrum of the output signal of the convolution integration means 30 is further emphasized in the band of large amplitude and the band of small amplitude is suppressed as compared with the spectrum of the input signal. However, the emphasized signal cannot be heard as sound because it is the signal after detection and integration. By normalizing the signal after the convolution integration with the signal before the convolution integration by the normalizing means 50, the gain of each band necessary for further enhancing the band of large amplitude and further suppressing the band of small amplitude is obtained. The multiplication means 60 multiplies the input signal divided into each band by the gain to emphasize or suppress each band, and finally the summing means 70 sums each band to obtain the output signal. Further, FIG. 10 shows a configuration diagram of another conventional voice processing apparatus. In FIG. 10, 100 is a buffer memory, and 110
Is FFT calculation means, 120 is power spectrum calculation means, 130 is convolution integration calculation means, 140 is function generation means, 150 is normalization means, 160 is multiplication means, 170
Is an inverse FFT means, 180 is a buffer memory, and 190 is a weighted addition means.

The operation of the speech processing apparatus of FIG. 10 configured as described above will be described below. For the sake of explanation, a case where a 2n-point real number type FFT is used in the FFT calculation means 110 will be described. The buffer memory 100 temporarily holds audio sampling data of 2n points or more. FFT
The calculation means 110 performs FFT. The power spectrum calculation means 120 calculates the power spectrum of the sum of squares of the real part and the imaginary part of the FFT calculation result. The convolution integration means 130, the function generation means 140, and the normalization means are shown in FIG.
However, the convolutional integration is performed every n sampling points in time. The multiplying means multiplies the real part and the imaginary part of the FFT result of the input signal by the gains of the respective frequencies obtained by the normalizing means 160 to further emphasize the band with large amplitude and suppress the band with small amplitude. Get the spectrum. The inverse FFT means 170 performs inverse FFT conversion on the output signal spectrum and outputs the time data at the 2n sampling points in which the band with large amplitude is further emphasized and the band with small amplitude is suppressed. The following process called overlap addition is performed in order to prevent a sudden change in gain at the boundary between adjacent FFT sections. The buffer memory holds the latter half n samples of the inverse FFT result for n sample times. The first half n samples of the inverse FFT result and the second half n samples of the previous inverse FFT result stored in the buffer memory and the weighted addition means 190 for each sampling ring time so that the weight of the new result increases with time. Is weighted and added.

FIG. 11 shows a block diagram of a conventional hearing aid. In FIG. 11, 210 is a low-pass filter, 220 is a band-pass filter, and 230 is a high-pass filter. Reference numeral 240 is a gain assigning means, 250, 260, 270 are multiplying means, and 280 is a summing means.

In the hearing aid configured as described above,
Low pass filter 210, band pass filter 22
0, the high-pass filter 230 divides the input signal into three bands of low band, middle band, and high band. Gain allocation means 240
Assigns the gain necessary for the spectrum of the input signal to fall within the audible range of the hearing impaired, depending on the frequency band of the input signal and the short-time average signal level. For example, for a hearing-impaired person with a narrow high-frequency range, assign a large gain to the high range when the high-range input signal is small, and put the signal in the audible range when the high-range input signal is small. Assign a small gain to and output. Multiplication means 250, 260, 2
At 70, the signals calculated in the gain allocating means 240 are multiplied by the signals in the respective bands, and the signals in the respective bands are amplified so as to be in the audible range of the hearing-impaired person. Finally, the output signal is obtained by calculating the sum of the signals divided into each band by the summing means 280.

[0012]

However, in the voice processing apparatus of FIG. 9, the gain is calculated at each sampling time, so the amount of calculation becomes huge, and a voice processing apparatus capable of processing voice in real time can be configured. It was difficult. In addition, if the formant enhancement effect is to be strengthened, it is necessary to increase the Q of each bandpass filter of the filter bank 10, which further increases the amount of calculation.

Further, in the voice processing apparatus of FIG. 10, since the gain is calculated once every plural sampling ring times, the calculation amount is small, but there is a problem that the formant cannot be emphasized sufficiently due to a large calculation error. It was Also,
When processing voice in a noisy environment. There is also a problem that annoying noise such as "curcules" is newly generated due to the fluctuation of the short-time spectrum of noise.

Further, with the hearing aid of FIG. 11, it was impossible to compensate for the deterioration of the frequency resolution of the hearing impaired person.

[0015]

The speech processing apparatus of the present invention comprises a filter bank for dividing an input frequency into a plurality of bands and a frequency analysis of the input speech for each of a plurality of sampling frequencies to obtain a high gain in a high energy band. Gain determining means for giving a low gain to the low energy band so as to correspond to each band of the filter bank, and multiplication for multiplying the output of the filter bank and the gain determined by the gain determining means for each sampling frequency. An output is obtained from the means and a summing means for summing the multiplication results of the first term.

The hearing aid of the present invention frequency-analyzes a first filter bank that divides an input signal into a plurality of bands, and performs a high gain in a high energy band and a low gain in a low energy band. Gain determining means for giving each band of the filter bank, and second dividing the input signal into a smaller number of bands than the first filter bank
Filter bank, gain allocation means for calculating a gain necessary for amplifying the signal after each division from the band-divided signal level divided by the second filter bank so as to be in the audible range of a deaf person, and gain determination. Multiplication means for multiplying the gain obtained by the means for each band corresponding to the gain obtained by the gain allocation means; multiplication means for multiplying the output signal of the multiplication means by the output signal of the first filter bank; The means sums up to obtain the output.

Further, the hearing aid of the present invention frequency-analyzes a first filter bank that divides an input signal into a plurality of bands, and a high gain is obtained in a high energy band and a low gain is obtained in a low energy band. Gain determining means for giving a gain corresponding to each band of the filter bank, first multiplying means for multiplying the output of the first filter bank by the output of the gain determining means for each band, and the input signal as the first A second filter bank for dividing into a smaller number of bands than one filter bank, and for amplifying the signal after each division from the band-divided signal level divided by the second filter bank so as to be in the audible range of a hearing-impaired person. A gain allocation means for calculating a necessary gain; a second multiplication means for multiplying an output of the gain allocation means and an output result of the first multiplication means for each corresponding band; and an output of the second multiplication means. Obtains and outputs the sum as the total means.

Further, the hearing aid of the present invention performs a frequency analysis on the input voice by dividing the input signal into a plurality of bands by a first filter bank, and a high gain is obtained in a high energy band and a low gain is obtained in a low energy band. Gain determining means for giving a gain corresponding to each band of the filter bank, first multiplying means for multiplying the output of the first filter bank by the output of the gain determining means for each band, and the input signal as the first A second filter bank for dividing into a smaller number of bands than one filter bank, and for amplifying the signal after each division from the band-divided signal level divided by the second filter bank so as to be in the audible range of a hearing-impaired person. Gain assigning means for calculating a required gain, first summing means for summing the outputs of the first multiplying means for each band of the second filter bank, and divided bands of the second filter bank. Doo and a second multiplying means for multiplying the outputs of the gain allocation means of the first summation means, for calculating and outputting the sum means the sum of the output of the second multiplier means.

Furthermore, the hearing aid of the present invention frequency-analyzes a first filter bank which divides an input signal into a plurality of bands, and performs a high gain in a high energy band and a low gain in a low energy band. Is provided so as to correspond to each band of the filter bank, first multiplication means for multiplying the output of the first filter bank and the output of the gain determination unit for each band, and several bands First summing means for calculating the sum of the signals, and gain assigning means for calculating the gain necessary for amplifying the signals of the respective sums from the output level of the first summing means so as to be in the audible range of the deaf person. The output of the first summing means is multiplied by the gain assigning means, and the sum of the output signals of the second multiplying means is calculated by the summing means and output.

[0020]

According to the present invention, it is possible to obtain a voice processing device having a high formant enhancement effect by the above-mentioned structure.

With the above-described structure, the present invention makes it possible to obtain a hearing aid that efficiently performs both formant enhancement and signal processing that puts the audio signal spectrum within the audible range of the hearing impaired.

[0022]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram of a voice processing apparatus according to a first embodiment of the present invention. In FIG. 1, 3
00 is a buffer memory, 310 is an FFT calculation means, 32
0 is a power spectrum calculation means, 330 is a convolution integration calculation means, 340 is a function generation means, 350 is a normalization means, 360 is a filter bank, 370 is a multiplication circuit, and 38
0 is a summing means. The 390 will be referred to as a gain determining means for the purpose of explanation below.

The operation of the speech processing apparatus of this embodiment constructed as above will be described. For the sake of explanation, a case where a 2n-point real number type FFT is used in the FFT calculation means 310 will be described. The buffer memory 300 temporarily holds 2n or more audio sampling data. FFT
The calculation means 310 performs FFT. The power spectrum calculation means 320 calculates the power spectrum of the sum of squares of the real part and the imaginary part of the FFT calculation result. The convolution integration means 330 convolves the average amplitude of each band with the function generation means 340 and the side suppression function on the frequency axis. In the output signal of the convolutional integration means 330, the large amplitude band is further emphasized and the small amplitude band is suppressed. The normalizing means 350 divides the signal after convolutional integration of each frequency by the signal before convolutional integration to determine the gain of each frequency for formant enhancement. These gains are
The input signal which is band-divided by 0 is multiplied, and the summation means obtains and outputs the summation of each band. Then, the high amplitude band is further emphasized and the low amplitude band is suppressed. FIG. 2 shows the enhancement effect of the figure formant of the vowel / a / uttered by a man. In FIG. 2, reference numeral 410 is the spectrum of the original sound, 420 is the spectrum of the processed sound by the sound processing apparatus of FIG. 1, and 430 is the spectrum of the processed sound by the conventional sound processing apparatus of FIG. Both processing devices used FFT of 128 points. As is clear from FIG. 2, the spectrum of the processed sound by the voice processing device of the present invention has a maximum of 30 in the band of small amplitude as compared with the original sound.
Although it is suppressed by about dB, it is suppressed, but the processed sound spectrum of the conventional sound processing apparatus has a small amount of suppression in a band of small amplitude.

As described above, according to the present embodiment, the FFT calculation means 310 is used to analyze the input voice, the gain is determined, and at the same time, the filter bank 360 is used to re-synthesize the voice. Compared with the case of re-synthesizing the voice by the inverse FFT, the contrast enhancement between the band where the energy is concentrated and the low energy band, that is, the formant enhancement of the voice can be performed more strongly. Moreover, since the gain is determined only once for every n sample points, the amount of calculation is small, and a voice processing device capable of voice processing in real time can be easily obtained.

FIG. 3 is a block diagram of a voice processing apparatus according to the second embodiment of the present invention. In FIG.
The same parts as those in FIG. 1 will be described with the same reference numerals. In FIG. 3, a frequency sampling filter 610 is added instead of the filter bank 360, the multiplication means 370, and the summing means 380, the buffer memory 620 and the weighted addition 630.
Is added, the high-frequency emphasis means 640 and the integration means 65.
The difference from FIG. 1 is that 0 is added. Further, the portion surrounded by the alternate long and short dash line is referred to as a gain determining unit 660 for the sake of later description.

The operation will be described below with reference to FIG. 3 having the above-mentioned configuration. In FIG. 3, the same reference numerals as those in FIG. 1 have the same functions as those in FIG. The output of the normalizing means 350 is updated once every n sample points. Adjacent FF
In order to prevent a sudden gain change from occurring at the boundary of the T section, the gain is smoothly changed at each sampling ring time by the following process called overlap addition. That is, the buffer memory 620 holds the previous gain calculation result. The newly calculated gain and the previous gain are weighted and added by the weighted addition means 630 so that the weight of the new gain increases with the passage of time. The sampling frequency of the frequency sampling filter 610 is designed to be the same as each frequency of the FFT spectrum. The input signal is applied to the frequency sampling filter, and the gain of each frequency, which is the output signal of the weighting and adding means 630, is applied as the gain of each sampling frequency at each sampling time to obtain a signal in which the formant is emphasized. For the purpose of explaining the high-frequency emphasizing means 610, FIG. 4 shows spectra of the vowel / a / before and after processing. In FIG. 4, 710 is the spectrum of the original sound, 7
Reference numeral 20 indicates a spectrum processed by the voice processing apparatus of the embodiment of FIG. 1, and 730 indicates a spectrum processed by the voice processing apparatus of the embodiment of FIG. The spectrum 720 has a high frequency and a maximum energy of 15 at the peak of the frequency as compared with the original sound.
It is attenuated by dB. This is perceived as an indistinct voice that is audible to the touch. Therefore, in order to improve the clarity, in FIG. 3, the high band of the gain is emphasized by the high band emphasizing means 610 after the gain is determined, and the sound whose peak of the spectrum is the same as the original sound from the low band to the high band is obtained. To get Further, when the voice processing apparatus of the embodiment of FIG. 1 is used in a noise environment, the FFT analysis time length is short, so that the frequency of the gain of the system corresponds to the fluctuation of the noise spectrum for a short time every n sample times. It fluctuates greatly and causes unnatural residual noise called "curcules". The integrating means 620 averages the gains over the FFT number interval,
Reduces unnatural residual noise called "curcules".

As described above, according to this embodiment, since the buffer memory 620 and the weighted addition means 630 are provided, deterioration of the naturalness due to the discontinuous change in the gain is prevented, and the high frequency gain enhancement means is provided. By doing so, a clear formant-enhanced voice can be processed, and by providing the integrating means 650, noise removing means that can remove unnatural residual noise can be obtained.

FIG. 5 shows a block diagram of a hearing aid in a third embodiment of the present invention. In FIG. 5, FIG.
The same parts as those in FIG. In FIG. 5, reference numerals 660 and 610 are the gain determining means and the frequency sampling filter described in FIG. 3, respectively. 810, 82
Reference numerals 0 and 830 denote a low pass filter, a band pass filter and a high pass filter, respectively. 840 is a gain allocation means. 850, 860 and 870 are multiplication means.

The operation of the hearing aid of this embodiment constructed as above will be described below. Gain calculation means 66
0 is the same as the gain calculating means in the embodiment of FIG. 1, and performs FFT every n sampling times and outputs the gain of each sampling frequency for performing formant enhancement every sampling time. On the other hand, the low-pass filter 810, the band-pass filter 820, and the high-pass filter 830 divide the input signal into three, the low-pass filter 810 has a low-frequency component, the band-pass filter 820 has a middle-frequency component, High pass filter 830 Outputs the time waveform of the high pass component. The gain allocating means 840 allocates the gain necessary for the spectrum of the input signal to fall within the audible range of the deaf person according to the frequency band of the input signal and the short-time average signal level. For example,
In the case of a hearing impaired person with a narrow high frequency range, when the high frequency input signal is small, a large gain is allocated to the high frequency range, and when the high frequency input signal is low, the signal is small enough to be in the audible range. Assign gain and output. In the multiplication means 850,
Of the gains for each sampling frequency for the formant enhancement calculated by the gain determining means 660, the low-pass filter 810.
Those in the band of are multiplied by the gain for the low frequency allocated by the gain allocation means 840, respectively. The multiplying means 860 and 870 also perform the same calculation for the mid-range and high-range frequencies, respectively. The frequency sampling filter 610 determines the frequency characteristic of the passing signal by the gain for each sampling frequency obtained as described above, and it is possible to put the spectrum of the signal in the audible range of the hearing-impaired person while performing formant enhancement. Becomes

As described above, according to the present embodiment, the gain determining means 660 for enhancing the formant and the input signal are divided into a plurality of bands so that the input signal is in the audible range of the hearing-impaired person. A gain allocating means 840 for allocating the gain of the band is provided, and by multiplying the gains determined by both to determine the gain of each frequency, formant enhancement is performed and the spectrum of the signal is placed in the audible range of the hearing-impaired person. It becomes possible to put in.

FIG. 6 shows a block diagram of a hearing aid according to a fourth embodiment of the present invention. In FIG. 6, the same parts as those in FIG. In FIG. 6, 10
Is a filter bank, 20 is a rectifying / integrating means, 30 is a convolution integrating means, 40 is a function generating means, 50 is a normalizing means, 60 is a multiplying means, and 70 is a summing means. These are conventional examples described in FIG. It has the same function as. The reference numeral 800 is referred to as a gain determining means for the purpose of explanation below. 810, 82
Reference numerals 0 and 830 denote a low pass filter, a band pass filter and a high pass filter, respectively. 840 is a gain calculation means, and 850, 860 and 870 are multiplication means.

The operation of the hearing aid of the present embodiment constructed as above will be described below. In FIG. 6, the filter bank 10 divides the input signal into a plurality of bands. The rectifying / integrating means 20 rectifies and integrates the band-divided signal for each band to obtain the short-time average amplitude. The convolutional integration means 30 convolves the average amplitude of each band with the side suppression function generated by the function generation means 40 on the frequency axis at each sampling time. In the output signal of the convolutional integration means 30, the large amplitude band is further emphasized and the small amplitude band is suppressed. However, the emphasized signal cannot be heard as sound because it is the signal after detection and integration. The normalizing means 50 divides the signal after the convolution integration in each band at each sampling time by the signal before the convolution integration to normalize and determines the gain of each band. The multiplication means 60 is the filter bank 1
The output signal of 0 is multiplied by the gain of each band which is the output of the normalizing means 50 to generate the output signal of each band. In these output signals, the large amplitude band is further emphasized and the small amplitude band is suppressed. On the other hand, the low-pass filter 810, the band-pass filter 820, and the high-pass filter 830 divide the input signal into three, and the low-pass filter 8
10 outputs a time waveform of a low frequency component, the band pass filter 820 outputs a time waveform of a middle band component, and a high pass filter 830 a high band component. The gain allocation means 840 is a frequency band of the input signal,
It assigns the gain necessary for the spectrum of the input signal to fall within the audible range of the hearing impaired, depending on the short-term average signal level. For example, for a hearing-impaired person with a narrow high-frequency range, assign a large gain to the high range when the high-range input signal is small, and put the signal in the audible range when the high-range input signal is small. Assign a small gain to and output.
In the multiplying means 850, of the output signals of each band calculated by the multiplying means 60, the low pass filter 81 described in FIG.
Those in the band of 0 are respectively multiplied by the gains for the low frequency allocated by the gain allocating means 840 and amplified so as to be in the audible range of the hearing-impaired person while being formant-emphasized. The multiplying means 860 and 870 also perform the same calculation for the mid-range and high-range frequencies, respectively. Finally, the summing means 70 calculates the sum of the output signals of each band and outputs the sum.

As described above, according to this embodiment, the output signal of the filter bank 10 is formant-emphasized by using the gain determining means 800 for the formant enhancement, and at the same time, the input signal is divided into a plurality of bands. The gain determining means 8 determines the gain of each band for allowing the input signal to enter the audible range of the hearing impaired person.
By stopping at 80 and multiplying both to obtain the sum, it becomes possible to put the spectrum of the signal in the audible range of the hearing-impaired person while performing the formant emphasis. In addition, since the input signal for formant enhancement is analyzed for each sampling frequency, it is possible to obtain a hearing aid that does not generate unnatural residual noise even when used in a noisy environment with little temporal fluctuation of the signal.

FIG. 7 shows a block diagram of a hearing aid according to a fifth embodiment of the present invention. 7, the same parts as those in FIGS. 1, 5 and 9 are designated by the same reference numerals for description. In FIG. 7, 10 is a filter bank, 60 is multiplication means, and 70
Are summing means, which have the same function as the conventional example described in FIG. Reference numeral 660 denotes a gain determining unit, which is shown in FIG.
It has the same function as. 840 is a gain allocation means, 85
Reference numerals 0, 860 and 870 are multiplication means, respectively, which have the same functions as those of the same numbers described in FIG. Reference numerals 1010, 1020, and 1030 are summing means for calculating the sum of signals of a plurality of bands input at each sampling time. Reference numerals 1040, 1050, and 1060 denote a low-pass filter, a band-pass filter, and a high-pass filter, respectively, and a low-pass filter 81, respectively.
0, band pass filter 820, high pass filter 8
It has the same pass band as 30.

The operation of the hearing aid of this embodiment constructed as above will be described below. Filter bank 1
0 divides the input signal into a plurality of bands. Gain determining means 6
As described in the description of FIG. 1, 60 calculates the gain of each band for formant enhancement. In the multiplication means 60,
The band-divided input signal is multiplied by the assigned gain of each band, the band with large amplitude is further emphasized, and the band with small amplitude is suppressed. On the other hand, the input signal is divided into the low-pass filter 810, the band-pass filter 820, and the high-pass filter 830, which divides the input signal into low-pass, middle-pass, and high-pass, and the gain assigning means 840 has been described with reference to FIG. In this way, the gain necessary for the signal in each band to be in the audible range of the hearing impaired is calculated and assigned to each band. The multiplication means is the summation means 101.
0, 1020, 1030 output signals and gain assigning means 88
A low-band, mid-band, and wide-band gain determined by 0 is multiplied to obtain a 3-band signal. Distortion due to non-linear amplification is removed by the low pass filter 1040, the band pass filter 1050, and the high pass filter 1060. Finally summing means 70
Then, the sum of the signals in the three bands input for each sampling time is obtained and used as the output signal.

As described above, according to this embodiment, the formant enhancement is performed in the band-divided state by the filter bank 10 and the gain determining means 660 for each band for the formant enhancement corresponding to the divided band. After that, the sum of the signals divided into the low band, the middle band and the wide band is obtained, and the gain allocation means 8 is provided.
By multiplying the gain of each band obtained in step 40, it becomes possible to put the spectrum of the signal in the audible range of the hearing-impaired person while performing the formant emphasis. Further, since the band division is performed by the filter bank, the sharpness of emphasis can be adjusted by the sharpness of resonance of the filter.

FIG. 8 shows a block diagram of a hearing aid according to a sixth embodiment of the present invention. In FIG. 8, the same parts as those in FIGS. 1, 5 and 9 are designated by the same reference numerals for description. In FIG. 8, 10 is a filter bank, 20 is rectification / integration means, 30 is convolution integration means, 40 is function generation means, 5
0 is a normalizing means, 60 is a multiplying means, and 70 is a summing means, which have the same functions as the conventional example described in FIG. Reference numerals 1010, 1020, and 1030 are summing means for calculating the sum of signals of a plurality of bands input at each sampling time. Reference numeral 840 denotes a gain allocating means, which allocates a gain necessary for the spectrum of the input signal to fall within the audible range of the hearing-impaired person according to the frequency band of the input signal and the short-time average signal level. 1110, 1120, 1130
Are multiplying means, which have the same functions as those of the same numbers described in FIG.

The operation of the embodiment of FIG. 8 configured as described above will be described below. In the band-divided signal in which the formants are emphasized by the same method as in FIG. 9, the band corresponding to the low band of the band gain allocating means 840 has a summing means 1010.
In addition, signals in each band corresponding to the middle range are summed up by the summing means 1020.
Then, each band corresponding to the high band is input to the summing means 1030, and output signals of each of the three bands after formant enhancement are obtained. Gains in each band necessary for putting the spectrum of this signal into the audible range of the hearing-impaired person are determined by the gain allocating means 840 and multiplied by the multiplying means 1100, 1120, and 1130, so that the voice after the formant enhancement is more reliably performed. It becomes possible to put it in the hearing range of the hearing impaired person.

As described above, according to the embodiment of the present invention, the summation means 1010, 1020, 1030 are provided to facilitate the band division of the low frequency band, the middle frequency band and the high frequency band, and the gain allocation means 840 determines the gain. By using the voice after the formant emphasis, it is possible to surely put the spectrum of the formant emphasized voice in the audible range of the hearing-impaired person.

The high frequency emphasizing device of FIG. 3 may be used not after the normalizing means but after the integrating means or after the frequency sampling filter. Further, in FIG. 3, the integrating means may be used after the power spectrum calculating means instead of after the weighting and adding means. Further, in FIG. 5, instead of the gain determining means 660, the gain determining means 390 described in FIG. 1 may be used. Further, in the embodiment of FIG. 6, the multiplying means 60 may be arranged after the multiplying means 850, 860, 870 to obtain the gain of each band and then multiply the band-divided signal.

[0041]

According to the present invention, it is possible to obtain a voice processing device capable of performing voice processing in real time and having a high formant enhancement effect. Unnatural residual noise, which was conspicuous in the conventional voice processing device, can be reduced.

Further, according to the present invention, it is possible to obtain a hearing aid that efficiently performs both the formant enhancement and the signal processing for putting the voice signal spectrum within the audible range of the hearing impaired person, and the implementation effect thereof is great.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a voice processing device according to a first embodiment of the present invention.

[Fig.2] Figure of vowel / a / uttered by a male Formant enhancement effect

FIG. 3 is a configuration diagram of a voice processing device according to a second embodiment of the present invention.

FIG. 4 is a spectrum diagram of a vowel / a / before and after processing.

FIG. 5 is a block diagram of a hearing aid according to a third embodiment of the present invention.

FIG. 6 is a block diagram of a hearing aid according to a fourth embodiment of the present invention.

FIG. 7 is a block diagram of a hearing aid according to a fifth embodiment of the present invention.

FIG. 8 is a block diagram of a hearing aid according to a sixth embodiment of the present invention.

FIG. 9 is a block diagram of a conventional voice processing device.

FIG. 10 is a block diagram of a conventional voice processing device.

FIG. 11 is a block diagram of a conventional hearing aid.

[Explanation of symbols]

300 Buffer memory 310 FFT calculation means 320 Power spectrum calculation means 330 Convolution integration calculation means 340 Function generation means 350 Normalization means 360 Filter bank 370 Multiplication means 380 Summation means 390 Gain determination means 410 Vowel / a / original sound spectrum 420 Vowels / a / spectrum of the processed sound by the voice processing apparatus of FIG. 1 430 vowel / a / spectrum of the processed sound of the voice processing apparatus of FIG. 10 610 frequency sampling filter 620 buffer memory 630 weighted addition means 640 high-frequency emphasis means 650 Integrating means 660 Gain determining means 710 Vowel / a / original sound spectrum 720 Vowel / a / processed sound spectrum by the voice processing apparatus of FIG. 730 Vowel / a / processed sound spectrum of the voice processing apparatus of FIG. 3 810 Low frequency Filter 820 band-pass filter 830 high-pass filter 840 gain assigning means 850 multiplying means 860 multiplying means 870 multiplying means 1010 summing means 1020 summing means 1030 summing means 1040 low-pass filter 1050 band-pass filter 1060 high-pass filter Band pass filter 1110 multiplication means 1120 multiplication means 1130 multiplication means

 ─────────────────────────────────────────────────── ─── Continuation of front page (72) Inventor Ryoji Suzuki 1006 Kadoma, Kadoma City, Osaka Prefecture Matsushita Electric Industrial Co., Ltd. (56) Reference JP-A-63-142399 (JP, A) JP-A-3- 266899 (JP, A) JP-A-4-24693 (JP, A) JP-A-4-25900 (JP, A)

Claims (12)

(57) [Claims] 1. A filter bank for dividing an input frequency into a plurality of bands and frequency analysis of an input voice for each of a plurality of sampling frequencies, wherein a high gain is provided in a high energy band and a low gain is provided in a low energy band. Gain determining means for giving each band of the filter bank, multiplying means for multiplying the output of the filter bank and the gain determined by the gain determining means for each sampling frequency, and summing means for summing the previous multiplication results. A voice processing device characterized by being provided. 2. A buffer memory for holding an input signal for an appropriate time, an FFT calculation means for reading the signal from the buffer memory and performing an FFT operation, a power spectrum calculation means for calculating a power spectrum from the FFT calculation result, and a specific A function generating means for generating a function, a convolution integral calculating means for convoluting the output of the power spectrum calculating means and the function generated by the function generating means, and a result of the convolution integration for each sampling frequency to the power. 2. The voice processing apparatus according to claim 1, wherein the gain determining means includes a normalizing means for dividing the output of the spectrum calculating means into a gain for each band. 3. A filter bank, a multiplication means for multiplying the output of the filter bank and the gain determined by the gain determination means for each sampling frequency, and a frequency sampling filter in place of the summation means for summing the multiplication results of the previous period. The voice processing device according to claim 1 or 2, characterized in that. 4. A buffer memory for temporarily holding the output of the normalizing means, and a past gain and a newly calculated gain held in the buffer memory for each frequency band are one sample during the time difference. 4. The voice processing device according to claim 1, further comprising a weighted addition means for performing interpolation in time so as to change smoothly with time. 5. The voice processing device according to claim 2, further comprising high-frequency emphasizing means for increasing the high-frequency gain of the gains of the respective bands determined by the normalizing means. . 6. The gain of each band determined by the normalizing means,
5. The voice processing apparatus according to claim 2, further comprising integration means for integrating each band with a specific time constant. 7. A first filter bank for dividing an input signal into a plurality of bands, and frequency analysis of an input voice to obtain a high gain in a high energy band and a low gain in a low energy band. Gain determining means for giving each band, a second filter bank for dividing the input signal into a smaller number of bands than the first filter bank, and a band division signal level divided by the second filter bank. Gain allocation means for calculating the gain necessary for amplifying the divided signal so as to be in the audible range of the hearing impaired person, and the gain obtained by the gain determination means is multiplied for each band corresponding to the gain obtained by the gain allocation means. A hearing aid comprising: a multiplying means, a multiplying means for multiplying an output signal of the multiplying means by an output signal of the first filter bank, and a summing means for summing the previous multiplication results. 8. A frequency instead of a first filter bank, a multiplying means for multiplying the output of the first filter bank and the gain determined by the gain determining means for each sampling frequency, and a summing means for summing the previous multiplication results. The hearing aid according to claim 7, further comprising a sampling filter. 9. A first filter bank for dividing an input signal into a plurality of bands, and frequency analysis of an input voice, wherein a high gain is provided for a high energy band and a low gain is provided for a low energy band. Gain determining means for giving each band, first multiplying means for multiplying the output of the first filter bank and the output of the gain determining means for each band, and an input signal less than that of the first filter bank A second filter bank that is divided into the number of bands and a band-divided signal level that is divided by the second filter bank are used to calculate a gain necessary for amplifying the signal after each division so as to be in the audible range of a deaf person. Gain allocation means, second multiplication means for multiplying the output of the gain allocation means and the output result of the first multiplication means for each corresponding band, and summing means for calculating the sum of the outputs of the second multiplication means. Hearing aid and said that there were pictures. 10. A first filter bank for dividing an input signal into a plurality of bands, and frequency analysis of an input voice to obtain a high gain in a high energy band and a low gain in a low energy band. Gain determining means for giving each band, first multiplying means for multiplying the output of the first filter bank and the output of the gain determining means for each band, and an input signal less than that of the first filter bank A second filter bank that is divided into the number of bands and a band-divided signal level that is divided by the second filter bank are used to calculate a gain necessary for amplifying the signal after each division so as to be in the audible range of a deaf person. The gain allocating means, the first summing means for summing the outputs of the first multiplying means for each band of the second filter bank, and the output of the first summing means for each divided band of the second filter bank. Second multiplying the output of the gain allocation means and
And a summing means for calculating the sum of the outputs of the second multiplying means. 11. A hearing aid according to claim 9, further comprising a band limiting filter corresponding to the band of the second filter bank after the second multiplication means. 12. A first filter bank for dividing an input signal into a plurality of bands, and frequency analysis of an input voice to obtain a high gain in a high energy band and a low gain in a low energy band. Gain determining means applied so as to correspond to each band, first multiplying means for multiplying the output of the first filter bank and the output of the gain determining means for each band, and the sum of signals for several bands is calculated. First summing means, a gain allocating means for calculating a gain necessary for amplifying a signal of each sum from the output level of the first summing means so as to be in an audible range of a deaf person, and a first summing means. Hearing aid, comprising: a second multiplication means for multiplying the output of the above by the gain allocation means, and a second summation means for calculating the sum total of the output signals of the second multiplication means.