JPH06332491A - Voiced section detecting device and noise suppressing device - Google Patents

Abstract

(57) [Summary] [Object] To obtain a voice section detection device and a noise suppression device which have a simple configuration and can be realized by a method that is small and easily consumes less power. [Structure] The output of a time constant circuit having an attack time of several tens of seconds and a release time of several ms is level-corrected by a multiplier to obtain a threshold value for voice section determination, so that it is easy, compact and low in power consumption. A simple voice section detection device is obtained. A noise suppressor is obtained by combining with an instantaneous compressor and a quasi-instantaneous compressor.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice section detecting device and a noise suppressing device.

[0002]

2. Description of the Related Art A conventional voice section detecting device is the simplest one that identifies a voice section exceeding a specific threshold value as a voice section. Assuming that this threshold is variable depending on the background noise level, Robert J. McAulay et al.
ch enhancement using a soft-decision noise supress
Ion filter "(IEEE ASSP-28 No.2 pp. 137-145) can be cited as a voice section detection apparatus. Fig. 6 shows a conventional voice section detection apparatus. In Fig. 6, 110 is an input signal. A level detecting means, a histogram generating means 120, and a voice section estimating means 130.

In the conventional voice section detecting device constructed as described above, the input signal level detecting means 110 calculates the average energy of the input signal every 20 ms per frame. The histogram generation means 120 generates a histogram for each input signal level devised so that the older frame data has a smaller influence based on the level of the input signal. Here, the generated histogram shows the cumulative distribution from the smaller energy of each frame up to several seconds before the present. The voice section estimation means 130 estimates the voice section based on the histogram. As an estimation method, when the cumulative frequency up to a specific amplitude value exceeds a threshold value, it is estimated as a voice section. Further, when it is not determined to be the voice section, the point at which the cumulative frequency is 80% is determined as the threshold value for dividing the noise section and the voice section.

As conventional noise suppression devices, a multiple microphone method, a maximum likelihood noise estimation method, etc. have been proposed. Among them, the multiple microphone method utilizes the fact that the difference in the intensity of the signal and the noise detected by the microphones installed at different positions is different for each microphone, and the signal from each microphone is multiplied by a constant coefficient and added. It suppresses noise. The maximum likelihood noise estimation method is "Speech enhanceme" by Robert J. McAulay et al.
nt using a soft-decision noise supression filter "
The noise suppression device used in (IEEE ASSP-28 No.2 pp. 137-145) can be mentioned. FIG. 7 shows a conventional noise suppression device.

In FIG. 7, reference numeral 310 denotes a voice section detecting device,
320a, 320b, ... are time constant circuits for obtaining the average amplitude of the noise section, 330a, 330b, ... are band division filters, 340a, 340b, ... are short-time energy calculation means, and 350a, 350b are maximum likelihood envelopes. It is a gain calculation means by estimation. 370a, 370b, ... Are multipliers and 380 is an adder.

The operation will be described below with reference to FIG.
The voice section detection device 310 discriminates the voice section in which the voice exists in the noise and the noise section in which only the noise exists for each 20 ms-long frame by the method used in FIG. on the other hand,
The input signals are band division filters 330a, 330b, ...
Is divided into a plurality of bands by the short time energy calculating means 3
40a, 340b, ... Calculates the energy within a 20 ms long frame. The time constant of the time constant circuit 320 is 1
Second, only the data of the frame estimated as the noise interval is input, and the weighted average energy of noise for several seconds is calculated. Gain calculation means 350 by re-likelihood envelope estimation
a and 350b calculate the gain Gn according to the following formula. Gn is small in a noisy band and is close to 1 in a low band. Multipliers 370a, 370b, ...
(1) multiplies the signals in each band by the assigned gain, and the adder 380 sums the signals in each band to obtain an output signal.

[0007]

[Equation 1]

[0008]

[Equation 2]

[0009]

However, in order to configure the voice section detecting device as shown in FIG. 6, the histogram generating means 120 and the voice section estimating means 130 having complicated functions are required, and the circuit scale is large. However, it has a problem in that it is difficult to be applied to a device such as a portable telephone because of its large size and power consumption. In view of the above point, the present invention is to realize a voice section detection device with a variable threshold value by a method that is simple, small in size, and low in power consumption.

Further, in order to construct the noise suppressing device as shown in FIG. 7, a gain calculating means by the re-likelihood noise estimation having a complicated function is required, the circuit scale becomes large, and the size and the power consumption are taken into consideration. It has a problem that it is difficult to adapt to a device such as a portable tape recorder. In view of the above points, the present invention provides a noise suppression device by a method that is simple, easy to reduce in size and low in power consumption.

[0011]

According to the present invention, an envelope extracting means for extracting an envelope of an input signal, an attack time connected to an output terminal of the envelope extracting means is 1 second or more, and a release time is 50 ms. The following first time constant circuit, a second time constant circuit connected to the output terminal of the envelope extraction means, having an attack time of 50 ms or less and a release time of 5 seconds or less, and a first time constant circuit. Output signal or second
The output signal of the first time-constant circuit after level correction is provided with a comparison means for level-correcting either of the output signals of the time-constant circuit of FIG. The voice section detecting device determines that the voice section is smaller than the output signal by the comparing means, and determines the voice section otherwise.

Further, according to the present invention, the envelope extracting means for extracting the envelope of the input signal, and the first connecting the output time of the envelope extracting means with an attack time of 1 second or more and a release time of 50 ms or less. And the attack time connected to the output terminal of the envelope extraction means is 50.
The level of either the output signal of the first time constant circuit or the output signal of the first time constant circuit or the second time constant circuit whose release time is 5 seconds or less in ms or less is level-corrected and then both are compared. Comparison means for comparing magnitude and first after level correction
If the comparison means determines that the output signal of the time constant circuit is smaller than the output signal of the second time constant circuit, the first gain is assigned; otherwise, the second gain is lower than the first gain. Is a noise suppressing device provided with a gain allocating means for allocating the gain of 1. and an amplifying means for amplifying an input signal with the gain allocated by the gain allocating means.

Further, according to the present invention, an envelope extracting means for extracting an envelope of an input signal, and a time constant connected to an output terminal of the envelope extracting means for an attack time of 1 second or more and a release time of 50 ms or less. A noise suppression device comprising a circuit and an instantaneous expansion means for instantaneously expanding an input signal, the expansion characteristic of which is determined by the output signal of the time constant circuit.

Further, according to the present invention, the envelope extracting means for extracting the envelope of the input signal, and the first connecting the output time of the envelope extracting means with an attack time of 1 second or more and a release time of 50 ms or less. And the attack time connected to the output terminal of the envelope extraction means is 50.
The characteristics are determined by the output signal of the second time constant circuit whose release time is 5 seconds or less in ms and the output signal of the first time constant circuit,
The noise suppressing device includes a gain assigning unit that determines a gain to be assigned based on an output signal value of the second time constant circuit, and an amplifying unit that amplifies an input signal with the gain assigned by the gain assigning unit.

[0015]

With the above-described structure, the present invention can realize the voice section detecting device in a simple and compact manner with low power consumption.

Further, according to the present invention, with the above-described configuration, the noise suppression device can be realized by a method which is simple, small in size, and low in power consumption.

[0017]

【Example】

(Embodiment 1) FIG. 1 is a block diagram of a noise suppressing apparatus using a voice section detecting apparatus according to a first embodiment of the present invention. In FIG. 1, reference numeral 510 is a voice section detecting device according to the present invention. Reference numeral 520 is a signal input terminal, and 530 is an envelope extraction means composed of a rectifier and a low-pass filter with a cutoff frequency of 70 Hz. 540 is a time constant circuit with an attack time of 30 seconds and a release time of 5 ms, and 550 is a time constant circuit with an attack time of 5 ms and a release time of 1 second. Reference numeral 560 is a level correcting means for multiplying the input by 5 and outputting it. Reference numeral 570 is a comparator. 580 is a gain allocation means,
590 is a multiplier, and 600 is an output terminal.

The operation of the noise suppressing device using the voice section detecting device of the present invention configured as described above will be described below. In FIG. 1, the envelope of the input signal is extracted by the envelope extracting means 530. The time constant circuit 540 outputs a value near the smallest value of the envelope for several tens of seconds. In a normal conversation, a voiceless portion is inserted every few seconds, so that a value in the vicinity of the minimum amplitude value of the voiceless portion is output. The time constant circuit 550 outputs a value near the short-term average amplitude. The level correction means 560 multiplies the output value of the time constant circuit 540 by 5,
The value with the smallest amplitude in the unvoiced part is corrected to a value near the average amplitude of the unvoiced part. This scaling factor is determined depending on the temporal dispersion of the noise envelope. This is the threshold for the voice section and the non-voice section. In the comparator 570, when the output signal of the level correction means 560 is smaller than the output signal of the time constant circuit 550, it is determined to be a voice section, and otherwise it is determined to be a non-voice section. The gain assigning means 580 assigns a gain of 1 to the voice section and a gain of 0.1 to the non-voice section according to the result of the comparator 570. The output signal is obtained by multiplying the multiplier 590 input signal by the assigned gain.
FIG. 2A shows a computer noise-superimposed speech waveform before processing and FIG. It can be observed that the voice section is estimated accurately and the noise section is suppressed.

As described above, according to this embodiment,
A voice interval detection device that is simple, compact, and easy to reduce power consumption by correcting the output of a time constant circuit having an attack time of seconds and a release time of several ms with a multiplier to obtain a threshold for voice interval determination Can be obtained.

(Embodiment 2) FIG. 3 is a block diagram of a noise suppressing apparatus according to a second embodiment of the present invention. In FIG.
The same parts as those in FIG. 1 will be described with the same reference numerals. Reference numeral 520 is a signal input terminal, and 530 is an envelope extraction means composed of a rectifier and a low-pass filter with a cutoff frequency of 70 Hz. 5
40 is a time constant circuit having an attack time of 30 seconds and a release time of 5 ms, and 600 is an output terminal. Reference numeral 710 is an instantaneous expander whose characteristics change according to the output value X of the time constant circuit 540, and the relationship between the input signal value i and the output signal value o is defined as shown in the figure. 720 is a low-pass filter.

The operation of the noise suppressing apparatus of the present invention configured as above will be described below. In FIG. 3, the envelope of the input signal is extracted by the envelope extracting means 530. The time constant circuit 540 outputs a value X near the smallest value of the envelope for several tens of seconds. In the decompressor 710, when i <3X, the input signal is completely suppressed because it has a high probability of being noise. When 6X ≦ i, it is determined that the section has a very good S / N, and the input is output as it is. 3X ≦ i <6
In the case of X, since it is a transitional section between the voice section and the noise section, the voice section and the noise section are smoothly connected by assigning outputs as shown in FIG. Low pass filter 7
Reference numeral 20 suppresses intermodulation noise outside the voice band generated by the non-linear amplification operation of the expander 710.

As described above, according to this embodiment,
The characteristics of the instantaneous expander are determined by the output of the time constant circuit having the attack time of seconds and the release time of several ms, and the input / output characteristics of the voice are determined, so that the configuration is simple, compact, and low power consumption is easy. Moreover, it is possible to obtain the noise suppression device in which the noise section and the speech section are smoothly connected.

(Embodiment 3) FIG. 4 is a block diagram of a noise suppressing apparatus according to a third embodiment of the present invention. In FIG.
The same parts as those in FIGS. 1 and 3 are designated by the same reference numerals for description. 52
0 is a signal input terminal, 530 is a rectifier and cutoff frequency 70H
It is an envelope extraction means composed of a low-pass filter of z. 540 has an attack time of 30 seconds and a release time of 5
ms time constant circuit, 550 has an attack time of 5 m
s, a time constant circuit with a release time of 1 second. 600 is an output terminal. 810 is the output value X of the time constant circuit 540.
It is a gain assigning means whose characteristic changes according to, and the relationship between the input signal value j and the output signal value o is defined as shown in the figure.

The operation of the noise suppressing apparatus of the present invention configured as above will be described below. In FIG. 3, the envelope of the input signal is extracted by the envelope extracting means 530. The time constant circuit 540 outputs a value X near the smallest value of the envelope for several tens of seconds. The time constant circuit 550 outputs a value j near the average amplitude for a short time. In the expander 710, j <3X
In the case of, the input signal is highly likely to be noise and is therefore completely suppressed. When 6X ≦ j, it is determined that the section has a very good S / N, and the input is output as it is. 3X ≦
When j <6X, it is a transitional section between the voice section and the noise section, so that the voice section and the noise section are smoothly connected by assigning a gain as shown in FIG. In the embodiment of FIG. 4, the gain is determined by the value j in the vicinity of the average amplitude for a short time as compared with the embodiment of FIG. 3, so that the intermodulation noise outside the voice band generated by the non-linear amplification operation of the gain allocation means 810. Becomes smaller. FIG. 5A shows a computer noise-superimposed speech waveform before the processing and FIG. The voice section is estimated accurately and the noise section is suppressed. Moreover, it can be observed that the noise section and the speech section are smoothly connected.

As described above, according to this embodiment,
The characteristic of the gain allocation means is determined by the output of the time constant circuit having an attack time of seconds and a release time of several ms,
By determining the gain based on the average amplitude in a short time, it is possible to obtain a noise suppression device having a simple structure, a small size, easy power consumption reduction, smooth connection between the noise section and the speech section, and less distortion.

In the embodiment of FIG. 1, the level correcting means 560 outputs the input signal by multiplying it by 5. However, other magnifications may be used or a specific value may be added. Instead of arranging the level correction means 560 at the subsequent stage of the time constant circuit 540, it may be arranged at the subsequent stage of the time constant circuit 550. Further, the time constants of the time constant circuits 540 and 550 used in the embodiments of FIGS. 1, 3 and 4 need not be the same as those of the embodiments as long as they meet the values defined in the claims. Needless to say. Further, in the embodiments of FIGS. 1, 3 and 4, the envelope extracting means is composed of the rectifier and the low-pass filter having a cutoff frequency of 70 Hz, but the envelope may be extracted by another method. Further, the characteristics of the instantaneous expander 710 and the gain assigning means 810 used in the embodiments of FIGS. 3 and 4 do not have to be the same as the figures, and differ depending on the temporal dispersion of the noise envelope to be suppressed. Needless to say, it is necessary to set it to a different characteristic. Further, the low-pass filter 720 may be removed in the embodiment of FIG. 3, or the low-pass filter may be attached to the output terminal of the embodiment of FIGS. 1 and 4.

[0027]

According to the present invention, a voice section detecting device having a variable threshold value can be realized by a method that is simple, small in size, and low in power consumption, and its practical effects are great.

Further, according to the present invention, the noise suppressing device can be realized by a method that is simple, small in size, and low in power consumption, and its practical effect is great.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a noise suppressing device using a voice section detecting device according to a first embodiment of the present invention.

[Fig. 2] Noise superimposed speech waveform diagram before and after processing

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

FIG. 4 is a configuration diagram of a noise suppressing device according to a third embodiment of the present invention.

FIG. 5: Noise superimposed speech waveform diagram before and after processing

FIG. 6 is a block diagram of a conventional voice section detection device.

FIG. 7 is a configuration diagram of a conventional noise suppression device.

[Explanation of symbols]

110 Input signal level detecting means 120 Histogram generating means 130 Voice section estimating means 310 Voice section detecting device 320a, 320b Time constant circuits 330a, 330b Band division filters 340a, 340b Short time energy calculating means 350a, 350b Gain by re-likelihood envelope estimation Calculation means 370a, 370b Multiplier 380 Adder 510 Voice section detection device 520 Signal input terminal 530 Envelope extraction means 540 Time constant circuit 550 Time constant circuit 560 Level correction means 570 Comparator 580 Gain allocation means 590 Multiplier 600 Output terminal 710 Instantaneous expander 720 Low pass filter 810 Gain assigning means

Claims (8)

[Claims] 1. An envelope extracting means for extracting an envelope of an input signal, and a first time constant circuit connected to an output terminal of the envelope extracting means for an attack time of 1 second or more and a release time of 50 ms or less. And a second time constant circuit connected to the output terminal of the envelope extraction means and having an attack time of 50 ms or less and a release time of 5 seconds or less, and an output signal of the first time constant circuit or a second time constant. The output signal of the first time-constant circuit after level correction is higher than the output signal of the second time-constant circuit after the level-correction of either of the output signals A voice section detection device, characterized in that if it is judged to be small by the comparison means, it is judged as a voice section, and in other cases it is judged as a non-voice section. 2. An envelope extracting means for extracting an envelope of an input signal, and a first time constant circuit connected to an output terminal of the envelope extracting means for an attack time of 1 second or more and a release time of 50 ms or less. And a second time constant circuit connected to the output terminal of the envelope extraction means and having an attack time of 50 ms or less and a release time of 5 seconds or less, and an output signal of the first time constant circuit or a second time constant. The output signal of the first time constant circuit after level correction is smaller than the output signal of the second time constant circuit, and the comparing means for level comparing one of the output signals of the circuit and then comparing the two. If the comparison means determines that the first gain is allocated, in other cases, the gain allocation means allocates a second gain lower than the first gain, and the gain allocated by the gain allocation means. input Noise suppression apparatus characterized by comprising an amplification means for amplifying No.. 3. An envelope extracting means for extracting an envelope of an input signal; a time constant circuit connected to an output terminal of the envelope extracting means, the attack time being 1 second or more and the release time being 50 ms or less; A noise suppression device comprising an instantaneous expansion means for instantaneously expanding an input signal whose expansion characteristic is determined by an output signal of a time constant circuit. 4. A first signal level and a second signal level higher than that are generated from the output signal of the first time constant circuit, and a zero value is generated when the input signal value is equal to or lower than the first signal level. When the value is above the second signal level, the output value of the first time constant circuit is output as it is, and when the value is between the first signal level and the second signal level, the instantaneous amplitude expansion is performed. 4. The noise suppression device according to claim 3, further comprising an instantaneous expansion means for performing the operation. 5. An envelope extracting means for extracting an envelope of an input signal, and a first time constant circuit connected to an output terminal of the envelope extracting means for an attack time of 1 second or more and a release time of 50 ms or less. A characteristic is determined by an output signal of a second time constant circuit having an attack time of 50 ms or less and a release time of 5 seconds or less connected to the output terminal of the envelope extraction means, and A noise suppressing device comprising: a gain assigning unit that determines a gain to be assigned based on an output signal value of a second time constant circuit; and an amplifying unit that amplifies an input signal with the gain assigned by the gain assigning unit. 6. A first signal level and a second signal level higher than that are generated from the output signal of the first time constant circuit, and the output signal value of the second time constant circuit is less than or equal to the first signal level. A gain of 0, a gain of 1 above the second signal level, and a gain of 0 to 1 for values between the first and second signal levels. 4. The noise suppressing device according to claim 3, further comprising a gain allocating unit. 7. The first signal level is defined as the ratio of the long-term average value and the minimum value of the envelope of noise to be suppressed multiplied by the output signal value of the first time constant circuit, and the second signal 7. The signal level is obtained by multiplying the ratio of the long-time maximum value and the minimum value of the envelope of noise to be suppressed by the output signal value of the first time constant circuit. The noise suppression device according to any one of claims. 8. The noise suppressing device according to claim 2, further comprising a low-pass filter for band-limiting the output signal.