JPH01101600A - Pitch cycle extraction - Google Patents

Abstract

PURPOSE: To reduce the misextraction of pitches and to accurately extract pitches by suppressing the generation of a double pitch and checking a period difference between the reference period waves of a current period. CONSTITUTION: The system is provided with a low pass filter(LPF) 7 and period difference judging devices 8 to 16. A voice is passed through the LPF 7 to suppress the generation of a double pitch to reduce the misextraction of pitches and a period difference between the reference frequency waves of a current period is checked to correct the misextraction of pitches. Therefore the generation of a double pitch is suppressed and a double pitch which is not suppressed is distinguished from an original pitch by checking a period difference between adjacent reference period waves, so that the influence of a double pitch is reduced and an accurate pitch period is extracted.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a pitch extraction method for an audio compression recording/playback device.

BACKGROUND OF THE INVENTION FIG. 2 shows the configuration of a conventional pitch period extraction method. In Figure 2, 1 is a peak holding circuit, the audio input is held in capacitor 2, and will not be charged unless there is an input with a voltage higher than the held voltage.At the same time, this voltage is due to the leakage current of resistor 8. , decreases exponentially. The reason why an exponential function is used here is due to the following circumstances.

The speech waveform within the pitch is approximated by the following equation.

y (large) = ΣAiexp (-g13it) cos (pr
fit-1-41) . . . Direct) fi: Formant center frequency 13i: y Half width ψi: Phase difference Ai: Formant amplitude level FIG. 8 shows an example of a speech spectrum. formant is 4
There are 4 to 6 formants in the range of KH2t, but as shown in FIG. 3, the amplitude of one or two formants in the low range is generally large, and in particular, the first formant determines the approximate periodic waveform.

Therefore, equation (1) is approximated by a trigonometric function whose envelope is an exponential decay waveform as shown in FIG. The time constant πB1 of this attenuation is determined by the formant half-width B+, and from analysis of many speech waveforms, B1 is approximately 20H2. The time constant RC of the decay waveform is adjusted to πB+.

Reference numeral 4 in FIG. 2 is a potential comparison control switch, which is turned on approximately 2 ms after being reset at the beginning of the pitch.

- This is because there is a secondary large amplitude component due to the second formant near the peak of the voice, which causes pitch extraction errors, so during the period where this secondary large amplitude component exists, This is to prevent the attenuation voltage from being compared with the audio input. 5 is a comparator between the attenuation voltage and the audio input voltage, and when the audio input voltage becomes higher, the counter 6 outputs a heliset signal. The counter 6 measures the interval between these resets, and the count value is sent to the observation system at the same time as the reset. This means that the pitch period of the voice is extracted as this count value.

FIG. 5 illustrates the principle of the present invention by drawing this attenuation waveform and the voice waveform on the same graph.

Problems to be Solved by the Invention: However, in the conventional pitch period extraction method described above, there is a problem in that when a high peak corresponding to the double pitch of the voice exists, it cannot be distinguished from the original pitch peak. The present invention is intended to solve such conventional problems, and aims to provide an excellent pitch period extraction method that can suppress double pinch peaks and make it possible to distinguish them from the original pitch peak. It is something to do.

Means for Solving the Problems The present invention provides a low-pass filter and a period difference determination device in order to achieve the above object.

Therefore, according to the present invention, the occurrence of double pitches can be suppressed by the low-pass filter, and the double pitches that cannot be suppressed can be distinguished from the original pitch by testing the period difference between adjacent fundamental period waves. This allows accurate pitch cycle extraction by reducing the influence of double pitch.

Embodiment FIG. 1 shows the configuration of an embodiment of the present invention. In Figure 1, 7 is a low-pass filter, 8 is an attenuated wave generation circuit from the peak, 9 is a comparator, 10 is a company counter,
11 and 12 are latch circuits, 13 is a subtracter, 14 is a threshold circuit for the result of subtraction, 15 is an AND circuit, and 16 is a counter.

Next, the operation of the above embodiment will be explained. In the above embodiment, the upper 2° of the low-pass filter 7 is cut by 4KH2 or more at about -48d1310Ct. This is because higher-order formants higher than the second formant, which cause double pitch generation, exist in the frequency range of approximately 2.4 KH2 or higher, so excluding this formant, the approximately periodic wave of the voice is as shown in Figure 4. , is intended to be a simple attenuation waveform. When this waveform enters the attenuated wave generating circuit 8, an attenuated waveform as shown in FIG. Let it start. However, when the second formant center frequency is low, the low-pass filter cannot suppress double pitch generation. Therefore, comparator 9
The counter 10. is synchronized with the pitch start signal output from the counter 10. The latch circuits 11 and 12 are operated to determine the periods of adjacent low-period waves, and the subtracter 18 determines the period difference, which is verified by the threshold circuit 14.

- Approximately 3.5 ms is used as the threshold value. This result is returned and ANDed with the pitch start signal in an AND circuit 15 to prevent an erroneous pitch start signal from being generated due to double pitch.

Effects of the Invention As is clear from the above embodiments, the present invention suppresses the occurrence of double pitches and tests the period difference between roughly periodic fundamental period waves to reduce erroneous pitch extraction. This has the effect of allowing accurate pitch extraction.

[BRIEF DESCRIPTION OF THE DRAWINGS] FIG. 1 is a circuit diagram of an apparatus for implementing a pitch period extraction method according to an embodiment of the present invention, FIG. 2 is a circuit diagram of an apparatus for implementing a conventional pitch extraction method, and FIG. FIG. 4 is a diagram showing a speech spectrum, FIG. 4 is a diagram showing a trigonometric function with an exponential attenuation envelope, and FIG. 5 is a diagram explaining the principle of a conventional pitch period extraction method. 7...Low pass filter, 8...Attenuated wave generation circuit,
9... Comparator, lO... Force' counter, 11,
12...Latch circuit, 18...Subtractor, 14...
-Threshold circuit, 15...AND circuit, 16...counter. 8th figure 3rd figure 4th figure W1! ■ Taste

Claims (1)

[Claims] A method of extracting pitch by comparing a speech waveform with a waveform that decays exponentially from the peak of a roughly periodic speech waveform.The method involves passing the speech through a low-pass filter, suppressing the occurrence of double pitches, and erroneously extracting the pitch. A pitch period extraction method that corrects incorrect pitch extraction by verifying the period difference between fundamental period waves of approximately periodicity.