JPH0619498A - Speech detector - Google Patents

Abstract

PURPOSE:To provide a speech which gives no feeling of physical disorder by detecting the head and tail of the speech as to the speech detector which detects a speech included in a pulse-code modulated speech. CONSTITUTION:After the DC component of the pulse code modulated speech which is inputted is passed through a high-pass filter 1 which cuts of the DC component, the output of the high-pass filter 1 is applied to a zero-cross quantity counter 2, an input electric power averaging part 3, and a predicted gain variation rate part 4 to calculate levels, frame by frame, and the calculation result is applied to a decision part 5 to detect whether or not there is the speech. This speech detector is provided internally with a sample power monitor part 6 which monitors the speech not in frame units, but in sample units after a no-sound decision made once, thereby detecting the head and tail of the pulse code modulated speech.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice detector for detecting voice contained in pulse code modulated voice.

In recent years, voice communication requires a method of compressing silence of voice and transmitting only voiced data in order to realize high efficiency of data. Therefore, there is a need for a voice detector that accurately detects the beginning of a voice and the end of a voice.

[0003]

2. Description of the Related Art A conventional example will be described below with reference to FIGS. FIG. 4 is a diagram showing the configuration of a conventional example circuit. Further, FIG. 5 is a diagram showing each detection timing of the conventional example circuit, and is a diagram for explaining the operation of FIG.

In FIG. 4, 1 is a high-pass filter,
By cutting off the DC component contained in the input pulse code modulated voice (PCM voice), the signal of the i-th harmonic component Si (i is the sample period number, where i = 1, 2, ... N) is output. It is what is output.

Reference numeral 2 denotes a zero-crossing counter, which calculates the number of zero-crossings for each frame of the i-th harmonic component Si output from the high-pass filter 1. The input power averaging unit 3 calculates the average power of the i-th harmonic component Si output from the high-pass filter 1 for each frame.

Reference numeral 4 denotes a predictive gain variation section, which is the predictive gain variation rate for each frame of the i-th harmonic component Si output from the high-pass filter 1, that is, ΣS during the entire sampling period.
The value of i × Si is calculated.

Reference numeral 5 is a determination unit, which is a zero-crossing number counter 2, an input power averaging unit 3, and a prediction gain fluctuation rate unit 3.
If the three output conditions are met, that is, if the three values reach a predetermined threshold value or more, the PCM voice is judged to be voiced, and if even one does not reach the predetermined threshold value, the PCM voice is output. It is determined that there is no sound.

Reference numeral 10 is a voice detector provided with the high-pass filter 1, the zero-crossing number counter 2, the input power averaging unit 3, the predictive gain variation unit 4 and the judging unit 5. As shown in FIGS. 4 and 5, it is assumed that the level of the signal Si is zero level in the area ~ t1 of the section ~, and the signal Si is substantially the threshold value in the area t1 ~.

In this case, any one of the three output conditions of the zero-crossing number counter 2, the input power averaging unit 3 and the predictive gain variation unit 4 becomes equal to or less than the threshold value during one frame, and thus one frame ends. At this point, the PCM voice is determined to be silent.

In addition, it is assumed that the signal Si has a certain constant level between the sections 1 and 2. In this case, the number of zero crossings, the average power, and the predicted gain fluctuation rate are equal to or higher than the predetermined threshold values, and therefore, the PCM voice is determined to be voiced at the end points and points of one frame.

Further, it is assumed that the signal Si is substantially a threshold value during the region ~ t2 of the section ~, and the level of the signal Si is zero level during the period t2 ~.
In this case, one of the three output conditions of the zero-crossing counter 2, the input power averaging unit 3 and the predicted gain variation unit 4 becomes equal to or less than the threshold value during one frame, and therefore, at the end point of one frame, The PCM voice is determined to be silent.

As described above, the number of zero-crossings, the average power, and the predicted gain variation rate of the PCM voice in the conventional voice detector 10 are calculated for each frame.
Therefore, even if the PCM voice becomes voiced at the end of the frame, the determination unit 5 may not be able to determine that the PCM voice is voiced.

[0013]

Therefore, in the conventional speech / tail detection method, there is a problem in that the user feels uncomfortable when he / she hears a speech whose speech / tail cannot be determined.

It is an object of the present invention to provide a voice that does not give an uncomfortable feeling by detecting the head and tail.

[0015]

In order to achieve the above object, as shown in FIG. 1, after passing through a high-pass filter 1 for cutting off a DC component of an input pulse code modulated voice, the high-pass filter 1 is passed through. By adding the output of the above to the zero-crossing number counter 2, the input power averaging unit 3, and the prediction gain fluctuation rate unit 4 to calculate each level for each frame, and the calculation result is determined by the determination unit 5.
In addition to the above, in the voice detector for detecting the presence or absence of voice, the voice detector is provided with a sample power monitoring unit 6 for monitoring the voice for each sample instead of the frame unit after the sound is once determined to be a pulse code. It is configured to detect the head and tail of modulated voice.

[0016]

In the present invention, as shown in FIGS. 1 and 2, after the PCM voice input in the sample power monitoring section 6 is judged to be voiced, the zero crossing number counter 2, the input power averaging section 3 and the predicted gain variation rate section are provided. 4 and the determination unit 5 are used to detect the voice at the beginning of the talk.

Also, in the tail end, even if the result of monitoring one frame by the zero-crossing number counter 2, the input power averaging unit 3 and the prediction gain variation unit 4 is judged to be silent, the sample power monitoring unit 6 is provided. Sounds are considered to be voiced.

Therefore, by using the presence / absence determination result, it is possible to detect the beginning / suffix with a reduced sense of discomfort in the voice.

[0019]

Embodiments of the present invention will be described in detail below with reference to FIGS. FIG. 2 is a diagram showing a configuration of an embodiment circuit of the present invention, and shows a circuit of an embodiment of a high efficiency voice codec for asynchronous modulation (ATM). FIG. 3 is a diagram showing each detection timing of the circuit according to the embodiment of the present invention.
FIG. 3 is a diagram for explaining the operation of FIG.

In FIG. 2, reference numeral 10 is a voice detector, which has a high-pass filter 1 and a zero-crossing counter 2 having the same structure as the conventional example.
An input power averaging unit 3, a predicted gain variation unit 4, a determination unit 5, and a sample power monitoring unit 6 of the present invention are provided. Further, 11 is a high-efficiency voice coding unit, and 12 is an ATM cell generation unit.

As shown in FIG.
The sample power monitoring unit 6 is used to monitor the PCM voice during one sample period. If the sample power monitoring unit 6 determines that the PCM voice is voiced in the region t1 to, the PCM voice in the interval to is accumulated in the determining unit 5 and the zero crossing number calculated in the next interval to. If the values of the average power and the predicted gain variation rate are voices equal to or higher than a predetermined threshold value, the PCM voice accumulated in the section ~ is sent from the voice detector 10 to the high-efficiency voice encoding unit 11.

In sections 1 and 2, the signal Si is above a certain threshold. Therefore,
The PCM voice is judged to be voiced by the voice detector 10. Further, the signal Si is equal to or more than a certain threshold in the areas 0 to t2 between the sections, but the signal Si is zero in the subsequent areas t2 to t2. Therefore, the voice detector 10 determines that there is sound in the areas 0 to t2, and after t2 that there is no sound.

As described above, the sample power monitoring unit 6
After judging the PCM voice with voice, the zero crossing number counter 2
By using the input power averaging unit 3 and the predictive gain variation unit 4, it is possible to reliably detect the presence / absence of voice.

On the other hand, in the case of the tail, even if it is judged that there is no sound when viewed in one frame, if the sample power monitoring section 6 has a sound, it is regarded as a sound. Further, the high-efficiency voice encoding unit 11 performs high-efficiency voice encoding only on the data determined to be voiced by the voice detector 10. And
The ATM cell generation unit 12 assembles the voice coded data from the high-efficiency voice encoding unit 11 into cells having fixed length packets for output to the ATM network, and sends the cells as cell output.

[0025]

As is apparent from the above description, according to the present invention, it is possible to detect the head and tail of a speech which has been determined to be silent by the conventional speech detector, and when the speech is decoded. The effect of being able to reduce the discomfort of is exhibited.

[Brief description of drawings]

FIG. 1 is a diagram showing a circuit of a principle configuration of the present invention.

FIG. 2 is a diagram showing a configuration of an embodiment circuit of the present invention.

FIG. 3 is a diagram showing each detection timing of a circuit according to an embodiment of the present invention.

FIG. 4 is a diagram showing a configuration of a conventional example circuit.

FIG. 5 is a diagram showing each detection timing of a conventional example circuit.

[Explanation of symbols]

 1 is a high-pass filter 2 is a zero-crossing number counter 3 is an input power averaging unit 4 is a prediction gain fluctuation rate unit 5 is a determination unit 6 is a sample power monitoring unit 10 is a voice detector

Claims (1)

[Claims] 1. A high-pass filter (1) for cutting off a DC component of an input pulse-code-modulated voice is passed through, and then the output of the high-pass filter (1) is compared with a zero-crossing counter (2) and an input power average. Section (3) and prediction gain fluctuation rate section (4) to calculate each level for each frame,
In addition to (5), in the voice detector (10) for detecting the presence or absence of voice, in the voice detector (10), a sample for monitoring the voice of each sample instead of the frame unit after the silence is once determined A voice detector characterized by being provided with a power monitoring unit (6) so as to detect the talk head and tail of a voice input.