JPS60216399A - Voice section detecting circuit for voice recognition equipment - 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 The present invention relates to a speech section detection circuit in a speech recognition device.

Generally speaking, when the target is input speech with a good signal-to-noise ratio, it is relatively easy to extract the section where speech exists. However, in an environment in which a speech recognition device is actually used, speech is input in a form that includes various noises and is superimposed on the noise. Since the noise at this time changes from moment to moment, it is difficult to detect stable voice sections by setting a fixed threshold value and cutting out voice sections, and this becomes a cause of misrecognition. When clipping is performed using such a fixed threshold, low-power parts such as the beginning and end of speech and voiceless consonants are cut out. Furthermore, under high noise conditions, noise will be added and cut out before and after what should originally be a voice section.

■--] The present invention was made in view of the actual situation mentioned in -.In particular, it detects stable speech sections regardless of the level of ambient steady noise, and achieves a stable recognition rate. This was done with the purpose of providing a voice section detection circuit that can ensure the following.

Configuration 13 of the present invention will be described below based on examples.

The feature of the present invention is to compare a preset high-level starting signal that is not affected by noise with a past signal that is delayed at a certain period, and determine whether the difference is constant and is greater than or equal to 2. is detected, and if it is greater than or equal to - room, it is further compared with the previous sample, and then the lowest threshold is determined, which is used as the optimal threshold, and the number of delayed samples is determined to be the minimum to detect the voice section. .

FIG. 1 is a time chart for explaining the operating principle of the present invention. In (a), A is the audio power signal, B is the optimal threshold, and n, n-1, n-2, etc. are samples. (b) shows the clock pulse, (e) shows the voice section signal, and (n-
1) Compare with the sample value at point, and if there is a difference greater than a set value, further compare with the sample value at point (n-2). When the difference is greater than the set value, it is further compared with point ('n-3) and similarly with point (n-4). For example, (n −3)
If it becomes less than the set value in comparison with point (n-4),
Section detection of the audio power signal is performed using the level at point (n-3) as an optimal threshold. In this way, the optimal threshold value can be set and the minimum number of delays can be obtained.

FIG. 2 is a diagram showing an embodiment of the voice section detection circuit according to the present invention. In the figure, 1 is a band pass filter (B P F
), 2 is a rectifier circuit, 3 is a low-pass filter ('LPF), 4 n is a 0-point comparator, 4 n
-1 is n-1 point comparator, 4n-2 is n-2 point comparator, 5 is sample hold circuit, 6 is falling edge detection circuit, 7 is threshold sample bold circuit, 8 is section 1
? n comparator, 9 is a low pass filter (L P F
), the input audio signal is sent to the BPFi, ll flow circuit 2.
．． It is converted into an audio power signal as shown by A in FIG. 1 through the LPF 3, and is input to a 0-point comparator 4n and a delay block hexagonal changeover switch Sn. In the initial state, selector switches Sn, 5n-1, old... are all ON.
The signal is sequentially delayed by the sample pole circuit 5 according to the clock pulse. The 0-point comparator 4n first compares the first n-point audio power signal with an arbitrary level to detect the starting point of the section signal. Then (n
-1) In the point comparator, the values of point n and point (n-1) are compared, and if the difference value is constant and is less than or equal to Bell, the output becomes i-level.

FIG. 3 is a diagram showing the configuration of a comparator for points below n-], in which comparator 10. The level difference between input signals No. 4-■ and ■ is taken, and the difference signal ◎ is compared with an arbitrary level ■ by the comparator 102. If it is below the set arbitrary level, Output becomes low level. Using such a comparator (n-
1) point and (n -2) point.

Compare point (n-2) and point (n-3)...
The exclusive OR of each output and the output of the comparator one time after it is taken, and the change sample point is detected.

This change sample point becomes the sample point from which the optimal threshold value is obtained.

At the change sample point obtained as described above, the delay sample and hold changeover switch is switched and input to the threshold sample and hold and the interval comparator. Also, detect the falling edge of each output of the previous exclusive OR signal, sample and hold the value of the previous delayed sample with that pulse, set it as the optimal threshold, and use it with the previous changeover switch. A section comparator compares the switched delayed voice power signal, which has already reached a fixed delay amount, to obtain a voice section signal. Also, since the delayed audio power signal is a discrete value, 'F?' is passed through LPF. It is possible to use the signal i') as the delayed voice power signal, and in this way, the delayed voice power signal (i') and the optimal voice section signal can be obtained.

(As is clear from the above description, according to the present invention, it is possible to detect a section signal by setting an optimal threshold value, and moreover, it is possible to automatically minimize the amount of delay.

[Brief explanation of the drawing]

Figure 1 is a diagram for explaining the operating principle of the present invention, Figure 2 is a diagram for explaining the operating principle of the present invention.
The figure is a diagram for explaining one embodiment of the voice section detection circuit according to the present invention, and FIG. 3 is a detailed diagram of a comparator having n-1 points or less. 1... Band pass filter (BPF), 2... Rectifier circuit, 3... Low pass filter (LPF), 4n...
・11 point comparator, 4n-1... n-1 point comparator, 4n-2... n-2 point comparator, 5...
- Sample hold circuit, 6... Falling edge detection circuit, 7... 7- Threshold sample hold circuit, 8... Section comparator, 9... Low pass filter (LPF), 101, 10
2...Comparator. −8=

Claims (3)

[Claims] (1) In a speech recognition device having a means for converting an input speech signal into a speech power signal, a means for generating a section signal from a preset threshold level of a speech section detection section, and means for comparing the n-time delay block and the threshold level with the level of (n-1); and the level of (n-1) with the level of (n-2).
, (n -2) level and (n -3),...
. A voice section detection circuit characterized in that it has a means for comparing up to once. (2) The delay amount of the delay block is variable and is automatically minimized.
1) The voice section detection circuit described in item 1). (3) In a speech recognition device having means for converting an input speech signal into a speech power signal, the speech section detecting section sequentially detects the corresponding thread from a point in time when the speech power becomes larger than a preset threshold level. It has a circuit that determines whether the level difference between the hold level and the previous one in the past is larger than a predetermined value, and if it is smaller, it compares the level difference between the previous one and the previous two times, Hereinafter, a voice section detection circuit is characterized in that the optimum section detection level is automatically detected by repeating similar comparisons.