JPS622300A - Voice pitch extractor - 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 Field of the Invention The present invention relates to a speech pitch extraction device that can be used for speech analysis and synthesis.

Conventional technology Conventionally, methods for determining speech pitch include the autocorrelation method, the modified autocorrelation method that uses the residual waveform obtained by linear predictive analysis of the speech waveform, and the absolute value calculation of the difference between the product-sum operation in the autocorrelation method. (hereinafter, functions for measuring waveform similarity, such as autocorrelation function, modified autocorrelation function, and average amplitude difference function, are collectively referred to as waveform similarity function method). (called the degree function).

However, these methods have the disadvantage that they require a large number of times of product-sum calculations and absolute-value sum calculations of differences, resulting in a long processing time. Therefore, in recent years, several methods have been proposed to reduce the processing time for speech pitch extraction. One of these methods is to detect the peak interval of audio waveform data and calculate the waveform similarity function only for a time shift width equal to this interval.

An example of the above-mentioned conventional audio pitch extraction device will be described below with reference to the drawings. FIG. 3 is a block diagram of main parts of a conventional speech pitch extraction device. In the same figure,
31 is a positive peak interval detection section, 32 is a waveform similarity function calculation section, and 33 is a pitch period determination section.

The operation of the conventional speech pitch extraction device configured as described above will be described below. When audio waveform data of a certain length of time is manually input to the positive peak interval detection section 31, a peak sequence having a value equal to or greater than a threshold determined by the maximum value of the positive sign of the waveform is detected, and the interval thereof is measured. Although positive peak intervals are detected here, negative peak intervals may also be detected.

Next, the detected peak interval is sent to a waveform similarity function calculating section, which calculates a waveform similarity function for a time shift width equal to this peak interval.

This waveform similarity function value is then sent to the pitch period determining section. Here, compare the waveform similarity function values for each time shift width, and select the time shift width that takes the maximum value when using an autocorrelation function as the waveform similarity function, and the minimum value when using the average amplitude difference function. Let it be the pitch period.

Problems to be Solved by the Invention In the speech pitch extraction device configured as described above, the sign of the peak interval to be detected is fixed to either positive or negative. However, when detecting peak intervals that are equal to or greater than a certain threshold, there is a large difference in the number of detected peaks between the positive sign side and the negative sign side of the waveform, depending on the shape of the audio waveform.

FIG. 4 shows an example where the threshold value is set to the maximum value.

In the case of the waveform (A), the number of peaks on the positive sign side is smaller than the number of peaks on the negative sign side (in the case of the waveform (B), the number of peaks on the negative sign side is smaller than the number of peaks on the positive sign side. Detecting many extra peak intervals other than the peak interval corresponding to the pitch period not only increases the computation time for calculating the waveform similarity function, but also causes pitch extraction errors.

The present invention has been made in view of this problem, and by selecting the polarity side with fewer extra peak intervals other than the peak intervals corresponding to the correct pitch period of the audio waveform data, more accurate audio pitch extraction is possible. The purpose is to provide equipment.

Means for Solving the Problems In order to solve the above-mentioned problems, the present invention counts and compares the number of positive and negative sign peaks of audio waveform data whose absolute value exceeds a threshold. a peak polarity selection section that outputs a peak polarity signal that selects whether to detect a peak interval with a positive sign or a peak interval with a negative sign according to a comparison value; and a variable polarity peak interval detection section that detects any of the peak intervals with a negative sign.

According to the above-described configuration, the present invention counts the number of positive and negative sign peaks of audio waveform data whose absolute value exceeds a threshold, and determines whether the positive sign side or negative sign side of the audio waveform data is positive or negative based on the magnitude relationship. It is determined which side of the negative sign has fewer extra peaks other than the peak series necessary for pitch extraction. This makes it possible to reduce the amount of calculation for calculating the waveform similarity function and to reduce pitch extraction errors.

Embodiment Hereinafter, a speech pitch extraction device according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram of essential parts showing an embodiment of the speech pitch extraction device of the present invention.

In FIG. 1, 11 is the positive sign of audio waveform data of a certain length that is digitized at a predetermined sampling period;
A peak that counts and compares the number of peaks for each negative sign whose absolute value exceeds a threshold, and selects whether to detect a positive sign peak interval or a negative sign peak interval depending on the comparison value. a peak polarity selection unit that outputs a polarity signal; 12 a variable polarity peak interval detection unit that detects either a positive sign peak interval or a negative sign peak interval according to the peak polarity signal; 13 a variable polarity peak interval detection unit that is equal to the peak interval; A waveform similarity function calculation unit 14 calculates a waveform similarity function value of the audio waveform data with respect to the time shift width, and 14 compares the waveform similarity function values and sets the maximum (or minimum) time shift width as the pitch period. This is a pitch period determination unit that outputs.

The operation of the voice pitch extraction device of this embodiment configured as described above will be described below. Note that the audio sampling frequency here is 8kHz.

The time length of the audio waveform data is 40m5. An average amplitude difference function is used as the waveform similarity function. 320 points of audio waveform data x (i) (i=1...
320) is input to the peak polarity selection unit 13, first, in the first half of the waveform, in the range of 1 to 160 points, the maximum peak value x (nA) of the positive sign of the waveform and the maximum peak value x of the negative sign of the waveform are input. (na) are detected, and each threshold value α=sc
*x (nA), β=k*x (n.

Set (0<k<1). Next, the number of positive peaks exceeding the threshold α over the range 1 to 320 is 1. And the number n- of negative peaks exceeding the threshold value β is counted.

If n, > n-, negative polarity is selected; if n+≦n-, positive polarity is selected; and the maximum peak position nA with a positive sign and threshold value α according to the peak polarity selection signal, or The maximum peak position nll with a negative sign and the threshold value β are output to the variable polarity peak interval detection section. Assuming that the pitch period search range is 2 to 20 m5 (16 to 160 points), the variable polarity peak interval detection unit 12 detects a positive peak position exceeding the threshold α over the range of nA+16 to nA+160 when the peak polarity selection signal is positive. Peak interval τ1 of maximum peak position nap. τ2... is output to the waveform similarity function calculation unit 13. Similarly, when the peak polarity selection signal is negative, the peak interval τ3. between the negative peak position exceeding the threshold β and the maximum peak position na over the range of nm +16 to na +160. τ2... is output to the waveform similarity function calculation unit 13.

), the waveform similarity function calculation unit 14 calculates an average amplitude difference function value expressed by the following equation for a time shift width equal to this peak interval.

Calculated average amplitude difference function value g(τ1)1g(τ2)・
... is input to the pitch period determining section 15.

The pitch period determining unit 15 detects the one having the minimum value from among the average amplitude difference function values, and outputs the time shift width as the pitch period.

FIG. 5 shows an example in which voice pitch is actually extracted.

The X mark is the result obtained by the conventional speech pitch extraction device using the method of detecting the peak interval of a positive sign, and the • mark is the result obtained by the speech pitch extraction device of the present invention. As can be seen from the comparison between the x mark and the * mark, it can be seen that the speech pitch extraction device of the present invention has fewer pitch extraction errors and is superior.

Effects of the Invention As described above, the present invention counts and compares the number of peaks of positive sign and negative sign of audio waveform data whose size exceeds a threshold value, and calculates the peak of positive sign according to the comparison value. a peak polarity selection section that outputs a peak polarity signal for selecting whether to detect an interval or a negative sign peak interval; By including the variable polarity peak interval detection section for detection, excellent effects can be obtained in that the amount of calculation for calculating the waveform similarity function can be reduced and pitch extraction errors can be reduced.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of the main parts of a voice pitch extraction device according to an embodiment of the present invention, FIG. 2 is a waveform diagram showing the operation of the variable polarity pitch interval detection section in the embodiment of the present invention, and FIG. 3 is a conventional FIG. 4 is a waveform diagram showing an example of a speech waveform, and FIG. 5 is a diagram showing an example of actual speech pitch extraction. 11... Peak polarity selection unit, 12... Variable polarity peak interval detection unit, 13... Waveform similarity function calculation unit, 14... Pitch period determination unit, 31・
. . . Positive peak interval detection section, 32 . . . Waveform similarity function calculation section, 33 . . . Pitch period determination section. Agent's name Patent attorney Toshio Nakao Baka I name Jl Toman ■ Whisper page 2

Claims (1)

[Claims] The number of positive sign and negative sign peaks of audio waveform data of a certain length of time digitized at a predetermined sampling period is counted and compared, and the positive sign is calculated according to the comparison value. A peak polarity selection unit outputs a peak polarity signal for selecting whether to detect a sign peak interval or a negative sign peak interval; a variable polarity peak interval detection unit that detects either;
A waveform similarity function calculation unit that calculates a waveform similarity function value of the audio waveform data for a time shift width equal to the peak interval, and a pitch that compares the waveform similarity function values and determines the maximum or minimum time shift width. 1. A voice pitch extraction device comprising: a pitch period determination section that outputs a pitch period.