JPS62294296A - Voice detector - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] 3. Detailed Description of the Invention Industrial Application Field The present invention is a voice transmission device for accurately detecting all voice signal sections from a signal mixed with noise, which is used in voice transmission devices and voice recognition devices. This is related to the number.

2. Description of the Related Art In recent years, speakerphones and teleconference systems have become widespread, and voice switches are used in these devices to prevent howling and eliminate indoor sounds. Furthermore, digitalization of audio transmission is progressing.
Research and development is also being conducted to improve transmission efficiency by distinguishing between non-voice sections and all voice sections, and transmitting signals only for the voice sections.

Furthermore, voice recognition technology, which uses human voice to control various devices, is rapidly progressing.

In order to improve the performance of these devices, there is a strong demand for a voice detection device that does not malfunction due to noise.

As a conventional technique of this speech recognition device, for example, the Journal of the Institute of Electronics and Communication Engineers, Vol. J63-Jin.

Fall 7. P, P, 159-176 (1980),
Institute of Electronics and Communication Engineers Technical Research Report, CAS84-238゜P,
P, 71-78 (1984), Iwatsu Giho, Vo
l.

201 Arashi i, p, p, 18-21 (1981), there is a voice detection device that reduces malfunctions caused by noise.

Hereinafter, the above-mentioned conventional voice detection device will be explained with reference to the drawings.

FIG. 7 shows a block diagram of a conventional voice detection device. In FIG. 7, 71 is a signal input terminal, 72 is a rectifier circuit that rectifies this signal, 73 is a maximum value hold circuit with a short rising time constant and a long falling time constant, and 74 is a falling time constant. 76 is a threshold value setting circuit that calculates the entire threshold signal from the output signal of this minimum value hold circuit 74; A determination circuit that compares the local maximum value signal obtained by the local maximum value hold circuit 73 and determines that an audio signal exists if the local maximum value signal is larger than the threshold signal; 77 is an output terminal for this determination signal. .

The operation of the voice detection device configured as described above will be described below.

FIG. 8 shows the signal waveforms at each part in FIG. 7, and the operation of FIG. 7 will be explained using this diagram.

In FIG. 8, 81 is the waveform of the input signal at point A in FIG. 7, 82 is the waveform of the rectified signal at point B in FIG. 7, 83 is the waveform of the maximum value signal at point 0, and 84 is the minimum value at point D. Waveform of value signal, 8
5 shows the waveform of the threshold signal at point E, and 86 shows the waveform of the determination signal at point F. First, the input signal 81 shown in FIG. 7 is input to the input terminal 71 shown in FIG. In this input signal 81, the period up to time To-T is a noise section, and the period from time 0 T1 to T2, in which no audio signal is included, is an audio signal section TEAL. Time T2 to T5 is a noise section.

As described above, the noise included in the audio signal includes electrical noise from equipment and noise due to sound absorption during audio recording. Noise noise is usually louder than electrical noise. Noise can be generated from air conditioning equipment, ventilation fans, automobiles, factory equipment, etc. The difference between noise caused by these types of noise and electrical noise is that its amplitude varies greatly over time. (The waveform shown in the input signal 81) Mainly contains noise due to obvious sounds, and the amplitude of the noise fluctuates greatly over time.

This input signal 81 is rectified by the frequent flow circuit 72 shown in FIG. 7 and becomes a rectified signal 82 shown in FIG. This rectified signal 82 is processed by a maximum value hold circuit 73 and becomes a maximum value signal 83.

On the other hand, the rectified signal 82) is also input to the return minimum value hold circuit 74, and becomes the minimum value signal 84.Next, this minimum value signal 84 is amplified by the threshold setting circuit 5, and the threshold value is The amplification factor of this threshold setting circuit 75 is set in advance so that the threshold signal 85 always has an extremely thick value of 83 or more in the noise interval from time To-T and from time T2 to T5. That is, the ratio between the maximum value and the minimum value of the noise amplitude is assumed in advance, and the amplification factor is set in accordance with this ratio.Next, the threshold signal 85 and the maximum value signal 83 are compared in the determination circuit 76. Here, when the maximum value signal 83 exceeds the threshold signal 85, it is determined that an audio signal is present, and a determination signal 86 is outputted from the output terminal 77 to an external device.This determination signal 86 is transmitted from time T1 to T
2, it is determined that it is ON, that is, a voice signal is present, and it can be seen that voice detection is performed accurately even under noise.

Problems to be Solved by the Invention However, the above configuration has the problem that when noise with large amplitude fluctuations is mixed into an audio signal, this noise is mistakenly detected as an audio signal. For example, if an air conditioner fan malfunctions and periodically generates high level noise, or if a printer or typewriter generates intermittent noise. In order to avoid erroneously detecting this noise as a voice signal, it is necessary to increase the amplification factor of the threshold setting circuit 75 and set the threshold value higher than the maximum value of the noise amplitude. However, with the conventional configuration, it was impossible to keep this amplification factor constant and control it to a threshold value by making it follow the noise with large amplitude fluctuations.

In view of the above-mentioned problems, the present invention makes it possible to control a threshold value by following the fluctuation width of the noise amplitude, and to perform voice detection that does not erroneously detect noise as an audio signal even if noise with a large fluctuation width is mixed in. It provides equipment.

problem? Means for Solving In order to achieve this objective, the voice detection device of the present invention comprises:
An amplitude calculation unit calculates the amplitude value by fully rectifying and smoothing the audio signal mixed with noise; a minimum amplitude calculation unit calculates the minimum amplitude value, which is the minimum value of this amplitude value; and a minimum amplitude calculation unit that calculates the noise amplitude in the noise interval from the amplitude value. A noise amplitude fluctuation width calculation section that calculates the fluctuation width, a threshold calculation section that calculates a threshold value from this noise amplitude fluctuation width and the minimum amplitude value, and a threshold calculation section that compares this threshold value and the amplitude value. The present invention includes a determination unit that determines that an audio signal is present when the amplitude value exceeds the threshold value.

With this configuration, the present invention can separate and detect the amplitude of unsteady noise that occurs intermittently as the noise amplitude fluctuation range, with the amplitude of stationary noise having a constant amplitude as the minimum amplitude value.

By calculating the threshold value from these two values, it is possible to set the optimal threshold value in accordance with the nature of the noise.

If stationary noise with a constant amplitude is the main noise, even if the threshold value is set close to the amplitude of the stationary noise, there is less risk of false detection and sensitivity can be increased. Conversely, if non-stationary noise is the main component, unless the threshold is set higher than the maximum amplitude of the non-stationary noise, when the noise becomes high it will be mistakenly detected as a voice signal. The voice detection device of the present invention can automatically control this threshold value, and can always perform voice detection with high sensitivity and without false detection.

EXAMPLE An example of the present invention will be described below with reference to the drawings.

FIG. 1 shows a block diagram of a voice detection device according to a first embodiment of the present invention.

In FIG. 1, 11 is a signal input terminal, 12 is an amplitude calculation unit that rectifies and smoothes this signal to calculate an amplitude value, and 13 is a minimum amplitude calculation unit that calculates the minimum amplitude value that is the minimum value of this amplitude value. , 14 is a noise amplitude fluctuation range calculation unit that calculates the entire noise amplitude fluctuation range in the noise interval, 15 is a threshold calculation unit that calculates a threshold value from this noise amplitude fluctuation range and the minimum amplitude value, and 16 is this noise amplitude fluctuation width calculation unit. This is an output terminal for a determination signal of a 17-bar determination unit that compares a threshold value with an amplitude value and determines that an audio signal is present when the amplitude value exceeds the threshold value.

Fig. 2 shows the signal waveforms at each part of the first embodiment? The operation of the first embodiment will be explained using this figure.

In FIG. 2, 21 is the waveform of the input signal at point A in FIG. 1, 22 is the waveform of the amplitude value at point B in FIG. 1, 23 is the waveform of the minimum amplitude value at point 0, and 24 ij: is the waveform at point D. 26 shows the waveform of the noise amplitude fluctuation width, 26 shows the waveform of the threshold value at point E, and 26 shows the waveform of the determination signal at point F.

First, the input signal 21 shown in FIG. 2 is input to the input terminal 11 shown in FIG. In this input signal 21, the period up to time To%T1 is a noise section and does not include a voice signal. The period from time T1 to time T2 is the audio signal section. Time T2-T3
is the noise section.

As explained in the conventional example, the waveform shown in the input signal 21 is an example in which the amplitude of the noise fluctuates over time. The amplitude value becomes 22 in Fig. 2. The input signal 21 is the result of addition of stationary noise with a substantially constant amplitude, non-stationary noise with a varying amplitude, and a voice signal. The minimum value of the amplitude values 22 indicates the amplitude value of stationary noise. To be careful, the amplitude value 22 is processed by the minimum amplitude calculation unit 13 to obtain the minimum amplitude value 23 which is the stationary noise amplitude. On the other hand, amplitude value 22i;
j, is also input to the noise amplitude fluctuation width calculation unit 14, and is input to the time To
%T1 and the peak and dip difference between times T2 and T5, the noise amplitude fluctuation width 24 is determined as the maximum amplitude value of the unsteady noise.
seek. Next, when the noise amplitude fluctuation width 24 and the minimum amplitude value 23 are inputted to the threshold calculation section 15, the threshold value 2
Calculate 5. This threshold calculation section 15 calculates the threshold according to the following equation.

(Threshold) = (Minimum amplitude value) + 2X (Noise amplitude fluctuation width) (1) Therefore, if the noise fluctuation is large, the threshold with a large margin is set, and if the noise fluctuation is small, the threshold The threshold value is automatically set to a value close to the minimum amplitude value with a small margin. Time 'ro,'
r1 and times T2 to T5, the threshold value 26 is the amplitude value 2
It is always set to a value higher than 2.

The determination unit 16 compares the threshold value 25 and the amplitude value 22, and if the amplitude value 22 exceeds the threshold value 2A, it is determined that an audio signal exists, and the determination signal 26 is determined. Output from the output terminal 17 to external equipment.

This determination signal 26 is turned ON between times T1 and T2, that is, it is determined that a voice signal is present, and it can be seen that voice detection is performed accurately even under noise.

Note that although equation (1) was used to calculate the threshold value, various methods can be considered for this calculation method. In the next example, which is generalized from equation (1) t, the threshold value can be controlled more finely using two constants K and C.

(Threshold value) = (minimum amplitude value) + Kx (noise amplitude fluctuation width) + CK>1, G>.

As described above, according to the present embodiment, the minimum amplitude calculation section 13 and the noise amplitude fluctuation width calculation section 14 calculate the amplitude of the non-stationary noise that occurs intermittently by setting the amplitude of the stationary noise as the minimum amplitude value. It is possible to separate and detect the noise amplitude fluctuation range. By calculating all the threshold values from these two values using the method of equation (1) in the threshold calculation unit 15, it is possible to set an optimal threshold value in accordance with the nature of the noise.

In other words, if the main noise is stationary noise with approximately constant amplitude,
Even if the threshold value is set close to the amplitude of this stationary noise, there is less risk of false detection and sensitivity can be increased.

Conversely, if non-stationary noise is the main component, unless the threshold is set higher than the maximum amplitude of the non-stationary noise, when the noise level becomes high, it will be mistakenly detected as a voice signal. In this embodiment, this threshold value can be controlled automatically, and a voice detection device that is always highly sensitive and does not cause false detections can be realized.

FIG. 3 shows a block diagram of a voice detection device according to a second embodiment of the present invention.

FIG. 3 shows more specifically the noise amplitude fluctuation range calculation unit 14 of the first embodiment, and the other configurations are exactly the same as the first embodiment.

In FIG. 3, 30 is a noise amplitude fluctuation range calculation unit, 31 is a short-time peak hold unit that calculates the envelope of the peak of the amplitude value, and has a fall time constant set to several seconds to several tens of seconds; 32 is a subtraction unit that outputs a subtracted value obtained by subtracting the minimum amplitude value from the envelope of this peak, and a minimum subtracted value calculation unit that calculates the minimum value of the subtracted value of 33, that is, the noise amplitude fluctuation width.

The other parts have exactly the same configuration as the first embodiment.

The operation of the voice detection device configured as described above will be described below.

FIG. 4 shows signal waveforms 2 at each part of the second embodiment, and the operation of the second embodiment will be explained using this diagram.

In Fig. 4, 41 is the waveform of the input signal at point A in Fig. 3, 42 is the waveform of the amplitude value at point B in Fig. 3, 43 is the waveform of the minimum amplitude value at point 0, and 44 is the peak at point D. The envelope waveform of
45 shows the waveform of the subtracted value at point E, 46 shows the waveform of the minimum subtracted value, that is, the noise amplitude fluctuation width, at point F, 47 shows the waveform of the threshold value at point 0, and 48i shows the waveform of the determination signal at point H.

First, the input signal 41 shown in FIG. 4 is input to the input terminal 11 shown in FIG. This input signal 41 is input to the amplitude calculation section 12 in FIG.
The signal is rectified and smoothed to obtain an amplitude value of 42 in FIG. Next, the amplitude value 42 is processed by the minimum amplitude calculation unit 13 to obtain the minimum amplitude value 43 which is the stationary noise amplitude.〇On the other hand, the amplitude value 42 is
The zero-order is also input to the noise amplitude fluctuation width calculation unit 30 and calculates the noise amplitude fluctuation width 46 as the maximum amplitude value of the unsteady noise.
When the noise amplitude fluctuation range 46 and the minimum amplitude value 43 are input to the threshold value calculation section 15, a threshold value 47 is calculated. The determination unit 16 compares the threshold value 47 with the amplitude value 42, and if the amplitude value 42 exceeds the threshold value 47, it determines that an audio signal is present and sends the determination signal 48 to the output terminal 17.
Output from to an external device. In this way, the basic operation is exactly the same as the first embodiment.

Next, the noise amplitude fluctuation range calculation unit 3 embodied in this embodiment
The details of the operation of 0 will be explained below.

First, the amplitude value 42 is input to the short-time peak hold section 31. The short-time peak hold section 31 has a falling time constant set to several seconds to several tens of seconds, and has an amplitude value of 4.
An envelope 44 of two peaks can be obtained. If we look at the peak envelope 44 in detail, we can see that it shows the maximum value of the noise amplitude in the noise section at time To%T. Its value increases rapidly in the audio signal section from time T to T2, but gradually attenuates in the noise section from time T2 to T3, and then reaches time T.
Around 5, the noise amplitude again reaches its maximum value.

Therefore, the subtraction unit 32 calculates the minimum amplitude value 43 from this value.
In the subtraction value 45 obtained by subtracting , the value from time To to T1 indicates the fluctuation range of the noise amplitude.

Also, the value in this interval is the smallest. Minimum subtraction value calculation unit 33
Then, the minimum value of this subtraction value 45 is determined and outputted to the threshold calculation section 15 as the minimum subtraction value, that is, the noise amplitude fluctuation width 46.

As described above, according to the present embodiment, the noise amplitude fluctuation width calculating section 3Q, which includes the short-time peak hold section 31, the subtracting section 32, and the minimum subtraction value calculating section 33, calculates the fluctuation width of the noise amplitude. can be determined accurately.

FIG. 5 shows a block diagram of a voice detection device according to a third embodiment of the present invention.

FIG. 5 shows another example of the configuration of the noise amplitude fluctuation width calculating section 30 of the second embodiment, and the other configurations are exactly the same as those of the first and second embodiments. , 5o is a noise amplitude fluctuation range calculation unit, 31 is a short-time peak hold unit whose falling time constant is set to several seconds to several Q seconds, and 51 is a peak envelope, which calculates the entire envelope of the peak of the amplitude value. A minimum peak calculation section 52 calculates the minimum value of the line, and a subtraction section 52 subtracts the minimum amplitude value from the minimum value of the envelope of this peak to obtain a subtracted value, that is, a noise amplitude fluctuation width. The other parts are 1st. This is exactly the same as the second embodiment.

The operation of the voice detection device configured as described above will be described below.

FIG. 6 shows signal waveforms in each part of the third embodiment, and the operation of the third embodiment will be explained using this diagram.

In Fig. 6, 61 is the waveform of the input signal at point A in Fig. 5, 62 is the waveform of the amplitude value at point B in Fig. 5, 63 is the waveform of the minimum amplitude value at point 0, and 64 is the peak at point D. The envelope waveform of
65 is the waveform of the minimum value of the envelope of the peak at point E, 66 is F
67 indicates the waveform of the threshold value at point G, and 68 indicates the waveform of the judgment signal at point H. The basic operation of this embodiment is as follows. This is exactly the same as the second embodiment. Therefore, the explanation will be focused on the operation of the noise amplitude fluctuation range calculating section 50, which is a feature of this embodiment.

First, the amplitude value 62 is input to the short-time peak hold section 31. This short-time peak hold section 31 is the same as that of the second embodiment. The short time peak hold section 31 has a falling time constant set to several seconds to several tens of seconds, and can obtain the envelope 64 of the peak of the amplitude value 62. Looking at the peak envelope 64 in detail, the time TO~
It can be seen that the maximum value of the noise amplitude is shown in the noise section of T1. The value increases rapidly in the audio signal section from time T1 to T2, but gradually attenuates in the noise section from time T2 to T5, and reaches the maximum value of the noise amplitude again near time T3. Therefore, is the minimum value 65 of the peak envelope and line calculated by the minimum peak calculation unit 51 the maximum value of the noise amplitude? show. The subtraction unit 32 subtracts the minimum amplitude value 63 from this value, and outputs it to the threshold calculation unit 15 as a noise amplitude fluctuation width jump.According to this embodiment, the short-time peak hold unit 31, a minimum peak calculation section 51, and a subtraction section 52, the noise amplitude fluctuation width calculation section 50 can accurately determine the fluctuation width of the noise amplitude.

Effects of the Invention The present invention uses a minimum amplitude calculation section and a noise amplitude fluctuation range calculation section to separate the amplitude of unsteady noise that occurs intermittently into a noise amplitude fluctuation range, with the amplitude of stationary noise as the minimum amplitude value. can be detected. By calculating the threshold value from these two values in the threshold calculation section, it is possible to set the optimal threshold value in accordance with the nature of the noise. If the main
Even if the threshold value is set close to the amplitude of this stationary noise, there is less risk of false detection and sensitivity can be increased.

Conversely, if non-stationary noise is the main component, unless the threshold is set higher than the maximum amplitude of the non-stationary noise, when the noise level becomes high, it will be mistakenly detected as a voice signal. According to the present invention, this threshold value can be controlled completely automatically, and a voice detection device that is always highly sensitive and does not cause false detection can be realized.

[BRIEF DESCRIPTION OF THE DRAWINGS] FIG. 1 is a block diagram of a voice detection device in an embodiment of the present invention, FIG. 2 is a waveform diagram of signals in each part of FIG. 1,
FIG. 3 is a block diagram of a voice detection device in a second embodiment of the present invention, FIG. 4 is a waveform diagram of signals in each part of FIG. 3, and FIG. 5 is a voice detection device in a third embodiment of the present invention. FIG. 6 is a block diagram of the device, FIG. 6 is a waveform diagram of signals in each part of FIG. 6, FIG. 7 is a block diagram of a conventional voice detection device, and FIG. 8 is a waveform diagram of signals in each part of FIG. 7. 12...Amplitude calculation unit, 13...Minimum amplitude calculation unit, 141 30.50...Noise amplitude fluctuation width calculation unit, 15...Threshold value Calculation Department, 16.
...Judgment section, 31... Short time peak hold section, 32... Subtraction section, 33... Minimum subtraction value calculation section, 61...・Minimum peak calculation part,
62... Subtraction section. 1st Ward Figure 2 Fff Cabinet Figure 4 ToT+ Tz Ts Time 6121 Ginma Figure 7 Fff 3 Figure 8 Th T+ Tz
TJ Kiran

Claims (3)

[Claims] (1) An amplitude calculation unit that calculates an amplitude value by rectifying and smoothing an audio signal mixed with noise, a minimum amplitude calculation unit that calculates a minimum amplitude value that is the minimum value of this amplitude value, and a noise interval from the amplitude value. a noise amplitude fluctuation width calculation section that calculates a fluctuation range of the noise amplitude at , a threshold calculation section that calculates a threshold value from this noise amplitude fluctuation width and the minimum amplitude value, and a threshold calculation section that calculates a threshold value from the noise amplitude fluctuation width and the minimum amplitude value; and a determination unit that compares the amplitude values and determines that an audio signal is present when the amplitude value exceeds the threshold value. (2) The noise amplitude fluctuation range calculation unit includes a short-time peak hold unit that calculates the envelope of the peak of the amplitude value, with a falling time constant set to several seconds to several tens of seconds, and a short-time peak hold unit that calculates the envelope of the peak of the amplitude value. The audio according to claim 1, characterized in that it is configured to include a subtraction section that subtracts the minimum amplitude value from and outputs the subtraction value, and a minimum subtraction value calculation section that calculates the minimum value of the subtraction value. Detection device. (3) The noise amplitude fluctuation width calculation unit includes a short-time peak hold unit whose fall time constant is set to several seconds to several tens of seconds, which calculates the envelope of the peak of the amplitude value, and a short-time peak hold unit that calculates the envelope of the peak of the amplitude value. Claim 1, characterized in that it is configured to include a minimum peak calculation section that calculates the minimum value of , and a subtraction section that subtracts the minimum amplitude value from this minimum peak value and outputs the subtracted value. Voice detection device.