JPS592918B2 - pitch extraction device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a pitch extraction device for extracting the fundamental pitch from a speech waveform.

In recent years, there has been a rush to develop voice processing devices such as voice recognition devices that can recognize voices and voice synthesizers that can synthesize voices, and one type of voice data used in these voice processing devices is As such, the basic pitch of speech is considered important.

The fundamental pitch of such a sound is the sense of pitch that one feels when listening to the sound, and corresponds to the lowest fundamental frequency among the five frequency components that form the sound.

Currently, there are 10 waveform processing methods that detect the periodic peaks of the waveform corresponding to pitch pulses seen in the voice waveform, and a method of extracting the basic pitch of the voice waveform after some preprocessing. There are methods such as a correlation processing method using an autocorrelation function or an autocorrelation function of a residual signal in linear predictive analysis.

Among such basic pitch extraction means, a pitch extraction device using waveform processing method 15 conventionally detects the presence of a peak by setting a predetermined threshold value for the peak of the audio waveform. Therefore, it was difficult to reliably detect the peak position of the fluctuating peak value of the audio waveform, and an accurate basic pitch could not be obtained.

The present invention has been made to solve the above-mentioned inconveniences, and provides a pitch extraction device that dynamically determines waveform peaks.

FIG. 1 shows the configuration of a pitch extraction device according to the present invention.

25 In the figure, 1 is a microphone that converts input voice into an audio signal, and 2 is a microphone amplifier that amplifies the audio signal from the microphone 1.

3 is an envelope detection circuit for detecting the envelope of the audio waveform of the audio signal obtained from the microphone amplifier 2;
An envelope extractor 4 that extracts the envelope of the audio waveform, a differentiating circuit 5 that differentiates the envelope signal of the audio waveform obtained from the envelope extractor 4, and whether the signal from the differentiating circuit 5 is positive. and a first comparator 6 that determines whether it is negative.

Reference numeral 7 denotes a first delay device that delays the audio signal obtained from the 35 microphone amplifier 2 by a predetermined time, and is used to ensure the detection time in the envelope detection circuit 3.

8 is a holding circuit that holds the maximum value of the delayed audio waveform obtained by further delaying the input audio waveform from the first delay device 7 for a certain period of time; The holder 10 stores the increasing peak value of the waveform every moment.

Reference numeral 11 denotes a first pitch circuit, and when the peak value of the input audio obtained from the first delay device 7 is larger than the maximum value of the delayed audio waveform held in the holding circuit 8, t? H1
? Based on the signals obtained from the second comparator 12 which outputs a level signal, or outputs an L level signal in the opposite case, and the first comparator 6 of the envelope detection circuit 3, the audio waveform is determined. and a first AND gate 13 which causes the signal from the second comparator 12 to be output when the envelope is increasing.

14 is a second pitch circuit, which includes an amplifier 15 that amplifies the peak value of the input audio obtained from the first delay device 7 at a constant amplification rate; If the peak value of the input audio waveform is greater than the maximum value of the delayed audio waveform held in the holding circuit 8, an IfHl level signal is output;
In the opposite case, the third comparator 16 that outputs the "7L" level signal and the signal obtained from the first comparator 6 of the envelope detection circuit 3 are inverted by the inverter 17, and the audio signal is inverted. and a second AND gate 18 which outputs the signal from the third comparator 16 when the envelope of the waveform is decreasing.

Incidentally, the amplification rate of the amplification device 15 is selected to be about 1.3, as will become clear later in the explanation. 19 is the second pitch circuit 14
It is an OR gate which takes the logical sum of the signal from the second AND gate 18 of the above and the signal from the AND gate 13 of the first pitch circuit 11 and outputs a binary signal of ゜1H'V level and WlL'1 level. This output represents the peak position of the input audio waveform.

Further, this output is also applied to the holder 10 of the holding circuit 8, and this holder 10 is operated only when the QlH'V level signal is output, making it possible for the holding circuit 8 to hold the maximum value of the delayed audio waveform. I have to. In the pitch extraction device of the present invention having such a configuration, when the audio output to the microphone 1 is in an increasing state, the envelope detection circuit 3 includes an envelope extractor 4, a differentiator 5, and a first comparator. Detecting the envelope of the input audio waveform that increases with 6? Outputs a 1H'1 level signal. Conversely, when the audio output to the microphone 1 is in a decreasing state, an Ln level signal is output. The operation of this envelope detection circuit 3 is similar to that of the first pitch circuit 11.
, is performed in advance of the operation of these circuits due to the delay effect of the first delay device 7 provided before the second pitch circuit 14 and the holding circuit 8. Therefore, when the audio signal input via the microphone 1, microphone amplifier 2, and first delay device 7 is increasing, the first The pitch circuit 11 becomes active, and the second pitch circuit 14 becomes disabled. Moreover, when the input audio signal is decreasing, the above is reversed. Next, when the input audio signal in an increasing state is introduced into the first pitch circuit 11, the periodic peaks Pl, P2, P3 corresponding to pitch pulses seen in the input audio waveform as shown in FIG.
A case will be explained in which the envelope A connecting the peak points of , increases with time.

In the same figure, the peak value of peak P1 is X1, and the time is t.
1. The peak value of peak P2 is X2, the time is T2, the peak value of peak P3 is X3, and the time when the peak value X of the input audio waveform becomes X=X1 is t'1 (however, t1<t'1
T2), and the time when x=X2 is t'2 (where T2
Let it be <t'2 × T3). Current time t is t1 t<t″1
At this time, X1 is held in the holding circuit 8, the peak value X of the input audio waveform becomes x<X1, and the first pitch circuit 11 compares the magnitude of x and The second
A low level signal as shown in FIG. b is output. When time t reaches t=t'1, X=X1 and beyond this point, the output 17H from the first pitch circuit 11? Switches to l level signal. This output is transmitted to the holding circuit 8 through the OR gate 19, and the holding circuit 8 is connected to this 11H
The acquisition of the increasing peak value of the delayed audio waveform is started by the 11 level signal. Time t is t″, t<.
At time T2, the peak value of the delayed audio waveform held moment by moment in the holding circuit 8 is smaller than the peak value x of the input audio which is increasing at this time, so the first pitch circuit 11 outputs 51H? Continue outputting l level signal. Time t =
At T2, the peak value of the input audio waveform is X&@x=X2,
When the peak value of the delayed audio waveform of the holding circuit 8 becomes approximately equal to this X2, and beyond this point, the first pitch circuit 11
The signal from T9L is switched to a T9L level signal. This one
Since the 1L1 level signal is transmitted through the OR gate 19 to the holding circuit 8, the holding circuit 8 stops taking in the peak value of the delayed audio waveform and enters a state where the peak value X2 is held. Furthermore, the time t becomes T2 t<,t'! When , the above-mentioned t1 x t<. Similarly to t'1, a ``QL6l level signal is obtained, and when t'2 is t<T3, the above-mentioned t''
1kut<. Similarly to T2, a V5H8 level signal is obtained. Ql obtained from the first pitch circuit 11 in this way
The Hl level and FlLl level signals become pulse signals that well represent the positions of periodic peaks seen in the audio waveform as shown in FIG. 2b. On the other hand, when the input audio signal in a decreasing state is introduced into the second pitch circuit 14, the periodic peaks Ql, Q2, Q3 corresponding to pitch pulses seen in the input audio waveform as shown in FIG. A case will be explained in which the envelope B connecting the peak points of decreases with time.

In the same figure, the peak value of peak Q1 is X1, and the time is t.
1. The peak value of peak Q3 is X2, its time is T2, the peak value of peak Q3 is t'1 (however, t1 x t'1<T2), α
The time when X=X2 is t'2 (where T2 x t'2<T3
). This α is the amplification rate of the amplifier 15 that constitutes the second pitch circuit 14 and amplifies the input audio signal, and is X1/α<X2 and X2/d<X3, and X1/α
And it is necessary that X2/α is larger than the maximum value other than peak Q2 and peak Q3. Under such conditions, this α is
It is appropriate to select it to be about 1.3. The current time is t1
When t<t″1, the holding circuit 8 holds X1, and the peak value X of the input audio waveform becomes x<X1/α,
The second pitch circuit 14 first multiplies x by d using an amplifier 15, and compares the magnitudes of αx and X1 with the third comparator 1.
6, and the comparison result is outputted as an L'1 level signal as shown in FIG. When time t is t-t″1, αX=X1
After this point, the second pitch circuit 1
The output from 4 switches to a 7H" level signal. This 1
! H! ? The level signal is transmitted to the holding circuit 8 through the OR gate 19, and the holding circuit 8 starts capturing the increasing peak value of the delayed audio waveform. When time t is t'1 and t<T2, the peak value held in the holding circuit 8 moment by moment is smaller than Dx, which is the amplification of the peak value X of the input audio at this time. The circuit 11 continues to output the "H8 level signal. When time t is t-T2, the peak value of the input audio waveform is the peak value X2 of the peak Q2, and beyond this point, the peak value of the input audio waveform is The value αX obtained by amplifying x becomes smaller than the peak value of the delayed audio waveform of the holding circuit 8, and the second pitch circuit 14
Switches to level signal. this! In response to the 1L1 level signal, the holding circuit 8 stops capturing the peak value of the delayed audio waveform and becomes in a state of holding the peak value can get. The pulse signal from the first pitch circuit 11 and the pulse signal from the second pitch circuit 12 thus obtained are outputted via the OR gate 19 as a series of pitch pulses.

As is clear from the above description, the pitch extraction circuit of the present invention includes a first pitch circuit that operates when the audio output is increasing, a second pitch circuit that operates when the audio output is decreasing, The first pitch circuit includes a comparison circuit that compares the maximum value of the delayed audio waveform and the audio signal, and the second pitch circuit compares the maximum value of the delayed audio waveform and the audio signal. Since it is composed of a comparison circuit that compares the amplification with the audio signal amplified at a constant amplification rate, the periodic peak corresponding to the pitch pulse seen in the audio waveform can be adjusted according to the fluctuation state of the peak value. The basic pitch of the audio waveform can be accurately obtained.

Moreover, the pitch information obtained from the pitch extraction device of the present invention is 2
Since it is a value pulse signal, an autocorrelation function based on this signal can be easily derived, and the pitch extraction device of the present invention is also optimal as a preprocessing means in a correlation processing method used in audio equipment. .

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of the pitch extraction device of the present invention, and FIGS. 2 and 3 are waveform diagrams for explaining the operation of the pitch extraction device of the present invention, in which 1 is an envelope detection circuit, and 8 is a holding circuit. circuit, 1
1 is the first pitch circuit, 12 is the second comparator, and 14 is the second pitch circuit.
and 16 a third comparator.

Claims (1)

[Claims] 1. In a pitch extraction device that extracts the basic pitch of a speech waveform, an envelope detection circuit that detects an envelope of the speech waveform;
a holding circuit that holds the maximum value of a delayed audio waveform obtained by delaying an input audio waveform for a certain period of time; a first pitch circuit that operates when the envelope detection circuit detects that the audio output is increasing; and a second pitch circuit that operates when the envelope detection circuit detects that the audio output is decreasing, and the first pitch circuit compares the maximum value in the holding circuit with the audio signal. The second pitch circuit comprises a comparison circuit that compares the maximum value in the holding circuit with an audio signal obtained by amplifying the audio signal at a constant amplification rate. A pitch extraction device characterized by the following configuration.