JPS628200A - Voice analysis system - 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] [Summary] The present invention relates to an audio processing method that inputs audio, analyzes it, and extracts information, and includes a reference point detection means for setting a fixed reference point in an audio signal, and a fixed amount. The input audio signals are sequentially stored in the storage means, and when a certain reference point is detected from the audio signal, the input audio signals are stored in the storage means for a certain period of time before. By starting the analysis from the data, it is possible to perform analysis synchronized with the reference point.

[Industrial Application Field] The present invention relates to a voice analysis method that inputs voice, converts it into a digital signal, and analyzes it, and particularly relates to a voice analysis method that can output stable analysis results for voice signal analysis. be.

[Prior art and problems] When analyzing and processing speech using a conventional speech analysis device,
The audio waveform, which is an analog waveform, is analyzed to extract various parameters.To do this, the analog audio waveform is extracted and held at a fixed sampling period, and the held values are sent to an A/D converter. After converting the digital data into digital data, a series of processes are performed on each data corresponding to an analysis frame cycle of a fixed time interval to extract various parameters.

In such conventional voice analysis methods, as soon as a signal is detected for a series of input voice signals, the data is simply processed and output at a fixed period, and the data is mechanically analyzed completely independently of the waveform of the voice signal. Since the extraction process is performed, there is a problem that even if signals having the same audio data are input, the data of each analysis frame is shifted, resulting in different analysis results, and that the analysis is time-consuming.

The present invention solves the above problems by performing analysis in synchronization with a certain reference point of the input audio signal. For example, it is possible to stably obtain the same analysis result for input signals of the same audio data. This provides a voice analysis method that can be used.

[Means for Solving the Problems] The means for solving the above problems include audio input means for digitizing audio signals, reference point detection means for detecting reference points from input signals, and and a voice analysis means for analyzing the signal held in the storage means at a constant frame period, and the reference point detection means converts the control signal into the voice analysis when detecting the reference point from the input signal. The voice analysis method is characterized in that the voice analysis means receives the control signal and starts analysis from data stored in the storage means a certain period of time before the reference point. 1 is a block diagram of the principle of the present invention, in which audio input means 1 A/D converts an input analog audio signal and sends it to storage means 3 and reference point detection means 2 as sampling data. The data is sequentially stored in the memory stage 3, and during this time the reference point detection means 2 detects a unique reference point from the data, and upon detection, sends a control signal to the voice analysis means 4. . The voice analysis means 4 receives the [69 signal, reads the data stored in the storage means 3 and corresponds to a certain period of time ago, and starts analysis.

[Example] 2nd rf! i is a block diagram showing an embodiment of the present invention;
The figure is 2nd rI! FIG. 4 is a diagram showing the relationship between the Kugata speech waveform and the analysis frame in the embodiment of I, and FIG. 4 is a time chart of the embodiment of FIG.

In the embodiment shown in FIG. 2, the input audio signal is sent from a microprocessor (hereinafter referred to as MPU) 5.
In accordance with the timing of LKI (shown in FIG. 4), the A/D converter 6 samples and A/D converts the data every sampling period Ts, and converts it into a digital value Xt (t indicates time). The digital value Xt is written into the variator memory 7 at the timing of CLK3 in FIG. 4 sent from the MPU 5. The address written at this time is the value indicated by the counter 8, and the value of the counter is counted up at the timing of CLKI as shown in FIG.
When the maximum address of the buffer memory 7 (255 in this example) is exceeded, counting starts again from 0. That is, Xo, X+, Xt-Xxs<, X2o,
The data x211,
・ will be written to each address. Therefore, data Xt is stored until the next data Xt, 2, and 6 are written at the same address.

In this embodiment, the reference point detection means includes a low-pass filter 9;
It is composed of a power calculation section 1o and a comparison section 11, and operates to detect the rise of a voiced sound from input data and use it as a reference point, and to send a control signal to the speech analysis section 12 when this is detected. The input data Xt passes through a low-pass filter having a cutoff frequency of 300 to 400 Hz to become data x't.

Next, the power calculation unit 10 calculates the short period power pt using the following formula.
Calculate.

pt=Σ(Xt-L>2 Next, the comparison unit 11 compares the calculated pt with a threshold for voiced sound determination, and sends a start signal to the voice analysis unit 12 when pt becomes larger than the threshold.

The voice analysis section 12 stops the analysis in response to the reset signal from the MPU 5, but starts the analysis upon receiving the start signal from the comparison section 11.

The analysis is a method of outputting the analysis results every fixed frame period Ta, and when the analysis is started, data is read from the buffer memory 7 at the timing of CLK2 in FIG. 4. The address of the buffer memory 7 at this time is the value indicated by the counter 8, and the data to be read out is at the time t.
Then, it is the data at t-256, that is, the Xt-t port. That is, as shown in FIG. 4, the input audio is A/D converted at the timing of CLKI to obtain Xlt, and
At the timing of CLK3, the data Xt-, .

Therefore, the data to be analyzed after receiving the start signal is data from a certain period of time before, depending on the capacity of the buffer memory 7. This relationship will be explained in the case where audio data as shown in FIG. 3 is input. The MPU 5 sends a reset signal to the voice analysis section 12 to stop the analysis.

When an audio signal as shown in the upper part of FIG. 3 is input, the data is A/D converted at every sampling period T3 and stored sequentially in the buffer memory 8. At time 0 t, the reference point S, which is the rise of the voiced sound, is detected. Then, a start signal is sent to the voice analysis section 12 to start analysis, but the data analyzed at this time is for a certain period of time Tm (Ts+=TsX 2
56 > The data at the previous point A is Xt-2ss, and thereafter the data is analyzed sequentially as Xt-2ss, Xt-ys4, etc., and the analysis results are output at every fixed frame period Ta, and a reset signal is sent from the MPU. Analysis will continue until it is sent.

[Effects of the Invention] As is clear from the above explanation, according to the present invention, it is possible to find a reference point specific to an input audio signal and perform analysis in synchronization with the reference point, thereby eliminating meaningless input. It is possible to efficiently analyze only the necessary parts without analyzing the signal, and moreover, for input data that is the same or similar, it is possible to stably and efficiently produce the same results without causing any discrepancies in the data of each analysis frame. It is extremely meaningful because it can produce good output.

[Brief explanation of drawings]

FIG. 1 is a block diagram of the principle of the present invention, FIG. 2 is a block diagram of an embodiment of the present invention, FIG. 3 is a diagram showing the relationship between input speech waveforms and analysis frames in the embodiment of the present invention, and FIG. It is a time chart of an example of the present invention.

Claims (1)

[Claims] audio input means for digitizing the audio signal; reference point detection means for detecting a reference point from the input signal; storage means for holding the input signal for a certain period of time; It has voice analysis means to analyze,
When the reference point detection means detects the reference point from the input signal, it sends a control signal to the voice analysis means, and the voice analysis means receives the control signal and detects the reference point from the reference point held in the storage means for a certain period of time. A speech analysis method characterized by starting analysis from previous data.