JPS60191300A - Voice section detecting circuit - 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 segment detection circuit in a speech recognition device, and more particularly to stabilization of segmentation of speech segments.

[In a speech recognition device, when input speech is targeted at speech with a good signal-to-noise ratio, it is relatively easy to extract sections where speech exists. However, the environment in which a speech recognition device is actually used includes various types of noise, and speech is input in a form mixed with noise. Since the noise at this time changes from moment to moment, it is difficult to detect a stable voice section using a method that sets a fixed threshold value and cuts out the voice section, which may cause misrecognition. Further, 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. Furthermore, under high noise conditions, noise is added before and after what should normally be a voice section and is cut out.

The present invention was made in order to solve the above-mentioned shortcomings of the prior art, and in particular detects a stable voice section regardless of the magnitude of the surrounding steady noise level. The purpose of this invention is to provide a speech section detection circuit that can ensure a high recognition rate.

1-Issei The configuration of the present invention will be described below based on examples.

The present invention is based on the assumption that the noise level between consecutively uttered discrete syllables or words does not change significantly between consecutively uttered syllables or words. Therefore, set the threshold for audio extraction by holding the peak value of the previous audio and dividing it by the peak or average value of the noise level before the start of the next audio to find the S/N at that time. The goal is to detect the beginning and end of audio as thresholds.

At this time, the point at which the noise level is detected is a time point delayed by a time T from the end of the previous voice. This time T is
Generally, in a word, there are 2 silent intervals such as the consonant /tsu/ in the word.
This is based on the fact that there are 00 to 400 m5, and the decision is made with the consideration that if the next voice is input within this range, the preceding and succeeding voices will be processed as one word (or unit).

FIG. 1 is a time chart for explaining the operating principle of the present invention, in which (a) is an example of the average signal level of the input signal, T, , T2 are the switching points of the threshold TH, A, B
, C, and D indicate the beginning (A, C) and end (B, D) of the audio, respectively. (b) shows a peak hold waveform, which is reset at point D, and indicates that the noise level and peak level are detected by the . and 0 marks. (C) shows the other peak hold waveform, which is similar to b), but has been reset at point B, and indicates that the peak level and noise level have been detected by the dot marks. (d) is the voice section detection pulse signal detected at the threshold (+U) in (a). (e) shows the flip-flop signal output Q that turns on at the falling edge of (d);
f) shows the reverse phase q of (e). In other words, this (e)
The rise and fall of the pulses (f) and (f) reset the peak hold circuit (see (b) and (c)). (g) is (
d) shows the waveform delayed by time T, and (h) shows the waveform of (g).
shows a flip-flop output that turns on at the falling edge of
This signal is used to switch the peak hold signal.

FIG. 2 is an electrical block diagram for explaining one embodiment of the present invention, in which l is an input section, 2 is a detection circuit, 3 is a smoothing circuit, 4 and 5 are peak hold circuits, and 6 1 is a level comparison circuit, 7 and 8 are switches, 9 and 10 are division circuits, 11 and 12 are flip-flop circuits, 13 is an output section, and the input signal from the input section l passes through the detection circuit 2 and the smoothing circuit 3. The average signal level is detected and input to peak hold circuits 4 and 5 and level comparison circuit 6.

Each peak hold signal is applied to one input terminal of a switch 7, 8, and the peak hold value is outputted from the switch 7, and the noise level is outputted from the switch 8 in synchronization. This signal is divided by S/N in a division circuit 9 to determine a threshold value, and is applied to a reference terminal of a level comparison circuit 6.

This division circuit 9 is linked to the switch 7.8 and holds the result of division at the time when the switch is turned on.
) repeats the reset and division hold at the falling edge.

The section signal shown in FIG. 1(d) is outputted from the comparator 6, guided to the output terminal 13, and simultaneously applied to the delay device 10 having a delay time T and the flip-flop circuit 12. The output of the delay device 10 is outputted to a flip-flop circuit 11 to obtain the output signal shown in FIG. 1(h), which controls the switch 7.8. On the other hand, the flip-flop circuit 12F7)Q, 'Clii No.
(See Figures (e) and (f)) are peak hold circuits 4 and 5.
The hold circuit is controlled as a reset signal. In this way, the threshold shown in FIG. 1(a) is set, and the section signal (FIG. 1(a)) is detected. The embodiment shown in Fig. 2 above uses the signal peak value and the noise peak value in the peak hold circuit to calculate the S/N, but for noise, the average value is used instead of the peak value. The S/N may be calculated from the peak value of the noise and the +F average value of the noise.

FIG. 3 is a diagram showing an example in which the S/N is calculated from the peak value of the signal and the average value of the noise. The same reference numbers as in FIG. 2 are provided. The difference between the embodiment shown in FIG. 3 and the embodiment shown in FIG. 2 is that the switch 8 used in the embodiment of FIG. The average signal level signal is directly applied to the signal. Similar to FIG. 2, this division circuit 9 holds the division value in conjunction with the switch. Note that the other operations are the same as those in Figure 2, so their explanation will be omitted.
In this embodiment, the threshold value 1a tends to be smaller than that in FIG.

Circle --] As is clear from the above description, according to the present invention, the noise level is detected between voices, and the previous voice level is determined based on the assumption that the level of the previous voice is the same as the next voice level. By determining the peak value of , the S/N ratio during this period, and using this as a threshold to detect the start and end of the next @ shore, it is possible to cut out a voice section in response to changes in both the noise level and the voice level. Therefore, more realistic section detection becomes possible.

[Brief explanation of drawings]

FIG. 1 is a time chart for explaining the operation of the present invention, and FIGS. 2 and 3 are electrical block diagrams for explaining the present invention in detail. DESCRIPTION OF SYMBOLS 1...Input part, 2...Detection circuit, 3...Smoothing circuit, 4.5...Peak hold circuit, 6...Level comparison port-path, 7.8...Switch, 9 .10... Division circuit, 11.12... Flip-flop circuit, 13.
...Output section. Figure 1 Figure 2 Procedural Amendment (Kiji) April 23, 1980 Patent Application No. 47940 2, Name of the invention Speech section detection circuit 3, Relationship with the person making the amendment Case Patent applicant Ota ri Nakamagome Address 1-3-6 Nakama-kome, Ota-ku, Tokyo Name (67,4) Ricoh Co., Ltd. Representative Ministry Hama 1)
Hiro 4, Agent Address: 7, Sha Train Yokohama 807, 1-2-7, Furo-cho, Naka-ku, Yokohama 231, Contents of amendment (1), "9 and 1O described in the second line of page 6 of the specification" is a division circuit," is corrected to "9 is a division circuit, and 1o is a delay circuit." (2) "r9, 10 river division circuit described in the second line of page 9 of the specification," was replaced with "9... division circuit, 10 river delay circuit,"
Correct to. (3) Correct Figure 1 as shown in the attached sheet. Figure 1

Claims (4)

[Claims] (1) In a speech recognition device, means for detecting the average signal level of an input signal, means for detecting and holding the peak value of this average signal level in parallel for each utterance unit, and detecting and holding the peak value of the average signal level for each utterance unit; means for switching between the peak value of the voice and the peak value of the noise; means for dividing the peak value of the signal by the peak value of the noise and holding the divided value; A voice section detection circuit characterized in that it has a means for comparing. (2) The voice section detection circuit according to claim (1), wherein the dark value is set after a certain time delay from the end of the voice. (3) In the speech recognition device, means for detecting the average signal level of the input No. 4g, means for detecting and holding the peak value of this average signal level in parallel for each utterance unit, and two peak value detection signals. means for calculating the peak value of the voice from , means for dividing the peak value of the signal by the average signal level of the noise and holding the divided value, and means for comparing the divided value with the average signal level as a dark value. What is claimed is: 1. A voice section detection circuit comprising: (4) The voice section detection circuit according to claim (3), wherein the threshold value is set with a certain time delay from the end of the voice.