JPS60101598A - Voice section 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 The present invention relates to a speech segment detection device, and more particularly to a word beginning detection device for stably cutting out speech segments in a speech recognition device.

In a conventional speech recognition device, when the target is speech with a good input signal-to-noise ratio, it is relatively easy to extract a section where speech exists.

However, in an environment in which a speech recognition device is actually used, speech is input in a form that contains various noises and is superimposed on the noise. The noise changes from moment to moment, and therefore, using a method of setting a fixed threshold value and cutting out a voice section, it is difficult to cut out a stable voice section, and this becomes a cause of misrecognition. As a solution to this
Various methods have been proposed for varying the threshold value, but each method has the drawback of being complicated and expensive.

-Comparison--The present invention was made in view of the above-mentioned circumstances.
In particular, the purpose of this invention is to provide a speech section detection device that can stably extract speech sections regardless of the level of ambient noise and ensure a stable recognition rate.

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

FIG. 1 is an audio signal waveform diagram for explaining the operating principle of the present invention, and FIG. 1(a) is a waveform diagram showing, for example, a change in the power of an input signal. Decided threshold T
This is the rough beginning of the word detected by H1. Threshold T in this case
H, is generally set to a value larger than the maximum value of the noise level. Figure 1(b) shows the signal in Figure 1(q) over time △
In the waveform diagram delayed by t, therefore, the point corresponding to point A is A'. The threshold value TH2 is a value set near the time point A is detected, and based on this value, an accurate rising point B is determined, and a correct speech section is determined. Therefore, it is sufficient for the device to start acquiring signals and detecting sections from the time point A is detected.

As described above, the present invention attempts to detect a speech interval by preliminary detection of the beginning of a speech using a fixed threshold value, and then detecting the exact beginning of the speech. 1 shows an electrical block diagram.

In FIG. 2, ■ is an input signal end, 2 is a delay device, 3 is a feature extractor, 4 is a quantization device, 5 is a voice section detector, 6 is a preliminary word beginning detector, and the input signal from input end 1 is is delay device 2
The output of the delay device 2 is applied to a preliminary word-initial detector 6, and the output of the delay device 2 is converted into a digital signal by a quantization device 4 such as an AD converter through a feature extraction unit 3 such as groups B, P, and F. A voice section is detected by a section detector 5 and output 8 as a section signal. Further, the detection signal 7 from the preliminary word-initial detector 6 is applied to the quantizer 4 and the speech section detector 5, and quantization and section detection are started based on this signal.

FIG. 3 is a detailed diagram of the preliminary word-initial detector 6 shown in FIG. is compared with the set threshold value TH of the threshold value setter 11 in the comparator 0, and the point A in FIG. 1(q) is detected.

However, the set threshold value is a value corresponding to a parameter such as power energy or zero-crossing wave number. The speech interval detector 5 is a detection unit that accurately detects the beginning of a word, and can use any of various known detection means, such as detecting the rise based on the difference between frames or detecting it from the power ratio of each band. It may also be detected by

FIG. 4 is a block diagram for explaining another embodiment of the present invention. In the figure, 12 is a shift 1 to register, 13 is a feature extractor, and other parts having the same function as those in FIG. are given the same reference numerals as in FIG. Therefore, in this embodiment, the delay device is not used for analog signals, but is configured with a shift register to handle quantized digital signals, and the input signal from input terminal 1 is processed by the quantization device. 4 and a preliminary word-initial detector 6. The signal quantized by the quantizer 4 is transferred to the shift register 12.
Then, a feature extractor 13 such as a digital filter extracts a feature parameter, and based on this parameter, a speech section detector 5 detects an accurate speech section 4t and outputs it 8. On the other hand, the detection signal 7 from the preliminary word-initial detector 6 is applied to the quantizer 4, the feature extractor 13, and the speech segment detector 5, and quantization, feature extraction, and segment detection are performed based on this signal.

As is clear from the above explanation, according to the present invention, the beginning of a word is roughly detected first, and then the beginning of the word is detected accurately. ) Since the beginning and section of the voice are detected, the section can be detected without false detection, and furthermore, the operation of the main body of the device can be limited to only after the preliminary beginning of the word has been detected, allowing efficient operation.

[Brief explanation of drawings]

FIG. 1 is a waveform diagram for explaining the operating principle of the present invention,
FIG. 2 is an electrical block diagram for explaining one embodiment of the present invention, FIG. 3 is a detailed electrical circuit diagram of the pre-word beginning detector 6 shown in FIG. 2, and FIG. FIG. 3 is an electrical block diagram for explaining another embodiment of the invention. 2... Delay device, 3... Feature extractor, 4... Quantization device, 5... Speech section detector, 6... Preliminary beginning detector, 10... Comparator, 11.・Threshold value setter, 12・
...Shift register, Engineering 3...Feature extractor. Figure 1 - (21-1

Claims (4)

[Claims] (1) It is characterized by having a detection means for preliminarily detecting the beginning of a word of a signal based on a preset threshold value, and a speech section detection means for detecting a detailed speech section based on the detection signal of the detection means. Voice section detection device. (2) The voice section detecting device according to claim 1, wherein a delay device is inserted before the voice section detecting means to delay the signal for a certain period of time. (3) The speech section detection device according to claim (1), wherein the preliminary word beginning detection signal operates the quantization means and the speech section detection stage. (4) The voice section detecting device according to claim 1, characterized in that a shift 1 register is inserted before the voice section detecting means to delay the signal for a certain period of time.