JPH0740200B2 - Voice section detection method - Google Patents

Description

The present invention relates to a voice section detection method in a voice confirmation device.

(Prior Art) Conventionally, in a voice recognition process, a voice section is detected. First, regarding the conventional voice section detection method,
It will be described with reference to the drawings.

According to the conventional voice section detection method, when the input level value of the voice signal is a function S (t) of the time t, the noise level value N when the voice signal is input, the input level value S (t), etc. The level threshold L is set more. A state in which the input level value S (t) is larger than the threshold value L (S (t)> L) is a constant time, that is, the starting end determination high level input minimum duration time TS.
When the above continues, the start time of this continuation time TS is set as the start end of the voice section. After that, when the state (S (t) ≦ L) in which the input level value S (t) is smaller than the threshold value L continues for a certain period of time, that is, at the termination determination low level input duration TE or more, the duration TE The voice section is determined by the determination method that the start time is the end of the voice section.

In this case, the noise level value L has a predetermined length from the time t0 when it is assumed that the voice signal is not being input,
Average value of the input level value S (t) in the noise measurement time interval TN that is continuous in time Are using. As the level threshold L, a method of setting a value L = N + C0 obtained by adding a predetermined constant C0 to the noise level value N is a general method.

According to this method, assuming that the time t1, t2, t3, t4 at which the level threshold L and the input level value S (t) match in FIG. 4, the input level value S (t) determines the level threshold L as described above. The start point of the section exceeding the duration TE of, for example, time t3 is determined to be the beginning of the voice section.

Next, the start point of the section below the level threshold L by the duration TE or more, for example, time t4, is determined as the end of the voice section.

(Problems to be Solved by the Invention) However, when the voice recognition device is actually used in an environment where noise fluctuations are severe, the input level value at the noise level value measurement time and a certain time from the noise level value measurement time It is generally considered that the probability that a value different from the input level value after the passage of time will increase with the passage of time. Therefore, the probability that the input level value (magnitude) at a time within a period where the time elapsed from the noise level value measurement time is relatively short, for example, immediately before the start of voice utterance, is different from the noise level value is relatively small. The probability that the start edge can be detected stably and accurately using the level threshold value set as in the prior art described above is high. However, the probability that the input level value (magnitude) at the time when the elapsed time from the noise level value measurement time is relatively long, for example, at the time immediately after the end of voice generation is different from the noise level value, is high. , The level threshold set at the beginning is not a value suitable for end detection, which increases the probability that the end cannot be detected accurately, which causes a problem that the voice confirmation performance deteriorates. there were.

The present invention has been made to solve the above-mentioned problems.

Therefore, an object of the present invention is to provide a method for detecting a voice section which enables stable and accurate detection of the end of the voice section.

(Means for Solving Problems) In order to achieve this object, according to the voice section detection method of the present invention, the following processing is performed (see FIG. 1).

According to the present invention, the temporary termination is determined as the first stage, and the true termination is determined as the second stage.

In this first step, the noise level value N of the input voice signal is smaller than the leading edge detection level threshold value LS obtained by adding a predetermined positive constant C1 and is larger than the noise level value N. The level value is obtained, and this level value is set as the temporary end detection level threshold value LKE. Next, the input level value S
(T) determines the start time t3 of the section in which the temporary end detection level threshold LKE falls below the predetermined temporary end determination low level input minimum duration time TKE as the temporary end of the voice section.

Next, in the second step, the magnitude of the input level value immediately after the provisional termination is used as the noise level value NE for detecting the termination, and a value obtained by adding a predetermined positive constant C3 to this noise level value NE is used for the termination detection. Let level threshold LE. Next, the stored input level value S (t) from the start end to the provisional end and this level value
Detected as the time t2 at which LE intersects and the true end.

(Operation) As described above, according to the present invention, the input level value S (t) after the provisional end detection value is used as the noise level threshold LE for detecting the true end.
Therefore, the end detection is not easily affected by the temporal fluctuation of noise, and therefore the detection of the voice section is stable and accurate.

(Embodiment) An embodiment of the voice section detection method of the present invention will be described below with reference to the drawings.

FIG. 1 is an explanatory diagram for explaining the voice section detecting method of the present invention, and shows an example of how the input level value changes by taking the time t on the horizontal axis and the level value on the vertical axis. FIG. 2 is a block diagram showing an example of the configuration of a voice section detection unit for carrying out the voice section detection method of the present invention. Further, FIG. 3 is a flow chart of the process used for explaining the method of the present invention.

The speech section detection unit shown in FIG. 2 includes a level extraction section 11, a temporary speech section detection threshold setting section 12, a temporary speech section detection section 13, a level storage section 14, a termination detection threshold setting section 15, and a termination detection section 16. And a control unit 17. In the following description, S represents a processing step in the flowchart.

First, the process is started (S1). The input voice signal a1 is input to the level detection unit 11, its level is extracted and converted into the input level signal a2 (S2). This input level signal
The input level value of a2 is S (t) and is shown by the solid line in FIG. The input level signal a2 is output to the temporary voice section detection threshold setting unit 12, the temporary voice section detection unit 13, and the level storage unit 14.

The control unit 22 outputs a temporary voice section detection threshold setting command signal r1 to the temporary voice section detection threshold setting section 12 at a time when it is assumed that no voice is being uttered. The temporary voice section detection threshold setting unit 12 receives the input level signal a2 for a predetermined time period TSN from the time t0 when the temporary voice section detection threshold setting command signal r1 is input, and the input level value in this time period. The average value of S (t) is the noise level value N, that is, And set. Next, a value obtained by adding a positive constant C1 determined by learning to the noise level value N in advance is determined as the start edge detection level threshold LS (S3).

LS = N + C1 The signal b1 of this level threshold LS is sent to the temporary voice section detection unit 13.

Next, in the temporary voice section detection threshold value setting unit 12, a positive constant determined in advance having a smaller value than the constant C1 used for setting the start edge detection belt threshold value with respect to the noise level value N. The value obtained by adding C2 is determined as the temporary end detection level threshold LKE (S4).

LKE = N + C2 This temporary end detection level threshold LKE is determined by adding a predetermined positive threshold value LS obtained by adding a predetermined positive constant C1 to the noise level value N of the input voice signal. Is subtracted from the constant C'2 to obtain a level value smaller than the level threshold LS and larger than the noise level value N, and this level value is used as the temporary end detection level threshold LKE.
May be set.

The temporary voice section detection threshold setting unit 12 sends the level threshold LKE signal b2 to the temporary voice section detecting unit 13 and sends the temporary voice section detecting threshold setting end signal b3 to the control unit 17.

When the temporary voice section detection threshold setting end signal b3 is supplied, the control unit 17 outputs the temporary voice section detection command signal r2 to the temporary voice section detection unit 13.

The temporary voice section detection unit 13, after receiving the temporary voice section detection command signal r2, starts the detection of the temporary voice section with the input level signal a2, the start edge detection level threshold b2 and the temporary end detection level threshold b3 as input, The start end and the provisional end are detected.

In the detection processing at the start time t1, the input level value S
From time t1 when (t) coincides with the start edge detection level threshold LS due to the passage of time, this input level value S (t) is the start edge determination high level input minimum duration time predetermined by learning.
If TS or more and a value greater than this level threshold LS S (t)> LS However, when the time period TS or more is detected, the above-mentioned start time t1 of this duration TS is determined as the start end of the voice section. (S5).

In the detection processing of the temporary end, the input level value S
This input level value S (t) is determined from the time t3 when (t) decreases the level after the start edge is detected and coincides with the temporary end detection level threshold value LKE, and the input end value S (t) is determined by learning, and the low level input minimum continuation is determined in advance. The value must be below the level threshold LS for time TKE or more S (t) ≤ LKE However, when the duration time TKE or more is detected, the above-mentioned start time t3 of this duration time TKE is temporarily terminated in the temporary voice section. (S6).

The signal d1 at the start end time t1 thus detected is output to the level storage unit 14, the end detection unit 16 and the control unit 17 (S5), and the signal d2 at the detected temporary end time t3 is output to the level storage unit 14 , End detection threshold setting unit 15, end detection unit 16, and control unit 17 (S6).

The signals d1 and d2 at the start time t1 and the temporary end time t3 are input to the level storage unit 14. When the start end time signal d1 is input, the storage of the input level value S (t) of the input level signal a2 is started from the start end time t1, and the storage of this input level value is started from the provisional end time t3 for a predetermined time predetermined by learning. Continue until the time that passes.

After receiving the provisional termination time d2 from the provisional voice section detection unit 13, the control unit 17 outputs the termination detection threshold value setting command signal r3 to the termination detection threshold value setting unit 15.

The termination detection threshold setting unit 15 receives the termination detection threshold setting command signal r3 from the control unit 17, and then the provisional voice section detection unit.
The input level value S (t) for a predetermined termination detection noise setting time TEN from the time t3 given by the provisional termination time signal d2 from 13 to the positive direction of the time axis is stored as the termination detection noise level signal e1. Received from Part 14. And
The average value NE of the input level values S (t) at this noise measurement time TEN, that is, Is set as the noise level value NE for end detection. Then, a positive constant C3 previously determined by learning is added to the noise level value NE to set the end detection level threshold LE (S
7). That is, LE = NE + C3 In this case, by selecting this constant C3, the end detection level threshold LE can be set.

The signal f1 of the level threshold LE is output to the end detection unit 16 and the end detection threshold setting end signal f2 is output to the control unit 17.

Upon receiving the end detection threshold setting end 16, the control unit 17 outputs the end detection command signal r4 to the end detection unit 16. After receiving the end detection command signal r4, the end detection unit 16 inputs the input level of the provisional voice section from the time t1 determined by the start end time signal d1 input from the provisional voice section detection unit 13 to the time t3 determined by the provisional end time signal d2. The signal e2 having the value S (t) is received from the level storage unit 14, and further the end detection level threshold signal f1 is received from the end detection threshold setting unit 15.

Then, in the end detection unit 16, these signals e2 and
The input level value S (t) of the temporary voice section determined by f1 and the end detection level threshold LE are compared in magnitude in the negative direction of the time axis from the temporary end time t3 to the input level value of the temporary voice section. The time when S (t) becomes larger than the end detection level value LE for the first time, for example t2, is detected as the true end time (S8).

The control unit 17 outputs the signal g at the end time determined in this way.
To the end of this voice section detection processing (S
9).

(Effects of the Invention) As is apparent from the above description, in the present invention, in order to cope with noise fluctuations when the speech recognition device is used, the end detection level threshold is set to the average value of the input level values immediately after the provisional end. It is based on. Therefore, since the end detection can be performed without being affected by the temporal fluctuation of the ambient noise level, stable and accurate voice section detection can be performed, and thus the recognition performance of the voice recognition device can be expected to improve.

[Brief description of drawings]

FIG. 1 is an explanatory view showing a state of temporal change of an input level value of an input voice signal, which is used for explaining an embodiment of a voice section detecting method according to the present invention, and FIG. 2 is an explanation of a voice section detecting method of the present invention. FIG. 3 is a block diagram showing a voice section detecting unit used in FIG. 3, FIG. 3 is a flowchart of processing of the voice section detecting method of the present invention, and FIG. 4 is an explanatory view of a conventional voice section detecting method. 11 …… Level extraction unit 12 …… Temporary voice section detection threshold setting unit 13 …… Temporary voice section detection unit 14 …… Level storage unit 15 …… End detection threshold setting unit 16 …… End detection unit, 17 …… Control unit.

Claims (1)

[Claims] 1. A noise level is defined as an average value of input level values of the input voice signal in a predetermined time period before the voice section of the input voice signal, and a level threshold value set based on the noise level, and the input. When detecting the voice section by comparing the level with the level value, the noise level value of the input voice signal is smaller than the level threshold for starting edge detection obtained by adding a predetermined positive constant and the noise level A level value having a value larger than the value is set as a temporary end detection level threshold, and the input level value falls below the temporary end detection level threshold by a predetermined temporary end determination low level input minimum duration time or more. The start time is determined to be the tentative end of the voice section, and from the tentative end, over a predetermined noise measurement period,
The average value of the input level value is used as the noise level value for end detection,
A value obtained by adding a predetermined positive constant to the noise level value is set as the end detection level threshold value, and the stored input level value from the start end to the provisional end,
A method of detecting a voice section, characterized in that a time closest to a tentative end is detected as a true end among crossing times with the level threshold for end detection.