JPH0378636B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a pitch extraction device that performs pitch extraction based on an autocorrelation coefficient obtained by analyzing a speech waveform with a frame period approximately equal to the pitch period. The present invention relates to a pitch extraction device that can significantly reduce pitch extraction errors in a section.

It is known that the voiced part of a speech waveform has a periodic repeating waveform, and the change characteristic of the period (pitch period) is an important parameter in speech analysis and synthesis, recognition, and the like. For example, in a speech analysis and synthesis system, the pitch extraction result extracted by the analysis section has a large effect on the quality of synthesized speech synthesized by the synthesis section.

Conventionally, the pitch period extraction method for audio waveforms is
Methods using various analysis parameters are known, such as a method of calculating and extracting an autocorrelation coefficient for each frame having a time length of approximately the pitch period.

Pitch extraction methods based on autocorrelation coefficients are widely used because the autocorrelation coefficients can be obtained through processing in the time domain, and the influence of the phase between the analyzed waveform and the frame is relatively small. . However, the pitch extraction method based on the autocorrelation coefficient often erroneously detects a period that is an integral multiple of the pitch period or N ₁ /N ₂ times the pitch period as the pitch period.
(However, N ₁ and N ₂ are integers, and N ₁ < N ₂ )
The above-mentioned problem in the pitch extraction method where a cycle that is an integer multiple of the original pitch cycle is mistakenly detected as a pitch cycle (hereinafter referred to as "integer multiple pitch cycle error") is addressed in the patent application proposed by the present inventor. Showa 54-
Detailed in 139307 "Pituchi Extraction Device".

Many pitch extraction devices based on autocorrelation coefficients limit the pitch search range to around the pitch period in order to alleviate integer multiple pitch period errors. Conventionally, in this type of pitch extraction device, a voiced sound section is detected using one voiced/unvoiced discrimination means, and a pitch search range is determined using the pitch period in the past voiced sound section. However, for voiced sounds,
This includes cases where the so-called pitch period extraction results are random, such as voiced plosives. The conventional method uses pitch period data in past voiced sound sections to determine the pitch search range.
This method has the drawback of determining an inappropriate pitch search range due to the influence of voiced plosive sections, resulting in pitch extraction errors.

An object of the present invention is to provide a pitch extraction device that is capable of stable pitch detection by removing data due to voiced plosives, etc. from past pitch cycle data used to determine a pitch search range. be.

A feature of the present invention is that voiced/unvoiced audio is classified for the purpose of determining the pitch search range, independently of the original voiced/unvoiced classification. In other words, the present invention eliminates data such as voiced plosives that are inappropriate as data for determining the pitch search range by specifying the voiced interval for determining the pitch search range using an independent voiced/unvoiced discrimination means. This has the effect of enabling more accurate pitch extraction.

Next, embodiments of the present invention will be described with reference to the drawings.

The figure is a block diagram for explaining an embodiment of the present invention. An input audio signal is input to an A/D converter 1. The A/D converter 1 converts the input audio signal into, for example,
Sample at 8KHz and quantize each sample. sampling,
The quantized audio signal is written into buffer memory 2. The buffer memory 2 outputs voice samples corresponding to one analysis window length, for example, 30 mSEC, to the voiced/unvoiced discriminator 3, the autocorrelation coefficient measuring device 4, and the voiced/unvoiced discriminator 5 every frame period, for example, every 20 mSEC. The voiced/unvoiced discriminator 3 determines whether the voice is originally voiced or unvoiced. Furthermore, the determination result is output to the pitch search device 6. The autocorrelation coefficient measuring device 4 measures the autocorrelation coefficient sequence of the audio waveform supplied from the buffer memory 2 for each frame period over a pitch period distribution range, for example, a delay range corresponding to 2.5 mSEC to 15 mSEC. Autocorrelation coefficient ρ of delay τ measured by autocorrelation coefficient measuring device 4
teeth It is. However, x(i) is the i-th audio sample in the analysis frame, and N is a natural number, for example, 15
Number equivalent to mSEC (720 for 8KHz sample)
It is. The measured autocorrelation coefficient sequence is supplied to a pitch searcher 6.

The voiced/unvoiced discriminator 5 performs voiced/unvoiced discrimination of the speech waveform supplied from the buffer memory 2 every frame period for the purpose of determining the pitch search range. The voiced/unvoiced discriminator 5 determines, for example, that the voice waveform supplied in each frame period is voiced if the power is greater than a predetermined threshold, and is unvoiced if it is smaller than the threshold. Furthermore, the determination result of the voiced/unvoiced discriminator 5 is outputted to the switch 7 as a control signal. The threshold value is set based on the fact that many stationary voiced parts have higher power than unvoiced sounds, unvoiced sounds, and voiced plosives, and its value is smaller than the average power of stationary parts of voiced sounds. The power is greater than the average power of a plosive sound. Therefore voiced/
The unvoiced discriminator 5 has the function of discriminating voiced stationary parts from other unvoiced sounds, unvoiced sounds, and voiced plosives.The pitch searcher 6 has the function of discriminating voiced stationary parts from other unvoiced sounds, unvoiced sounds, and voiced plosives. The maximum value in is searched, and the delay time τ _P corresponding to the maximum value is determined. The pitch search range is supplied from a pitch search range measuring device 9. The pitch searcher 6 is also voiced/
If the voiced/unvoiced discrimination result supplied from the unvoiced discriminator 3 is voiced, the τ _P is output to the pitch output terminal 10 and the switch 7, and if it is unvoiced, the τ _P is changed to "0" and the pitch is output. Output to output terminal 10 and switch 7. In other words, voiced/unvoiced discriminator 3
The method discriminates voiced parts including voiced plosive parts, and the well-known method described above is sufficient for the discrimination method.

The switch 7 switches between τ _P and "0" in response to a control signal supplied from the voiced/unvoiced discriminator 5, and further outputs the signal to the temporary memory 8. When the control signal indicates voiced, the switch 7 sets τ _P ;
When indicating silence, "0" is output to the temporary memory 8. The temporary memory 8 sequentially stores τ _P or "0" supplied from the switch 7 every frame period. The temporary memory 8 outputs the pitch period τ _P of the stored past frame to the pitch search range measuring device 9 for each frame period.

The pitch search range measuring device 9 measures the pitch search range from the pitch period of past frames and outputs the result to the pitch search device 6. The method of determining the pitch search range is a conventional method, for example, the method mentioned in the above-mentioned Japanese Patent Application Laid-Open No. 54-139307 and Japanese Patent Application Laid-Open No. 56-42296.

The formula for calculating ρ〓 in the autocorrelation coefficient measuring device 4 is, for example, Or it can be simplified to τ= _N 〓 ⁱ⁼¹ x(i)×x(i+τ).

As described above, according to the present invention, when determining the pitch search range, it is possible to remove the influence of voiced plosives and the like whose pitch period is random, and it is possible to extract pitches with high accuracy.

[Brief explanation of drawings]

The figure is a block diagram for explaining an embodiment of the present invention. 1...A/D converter, 2...Buffer memory,
3... Voiced/unvoiced discriminator 1, 4... Autocorrelation coefficient measuring device, 5... Voiced/unvoiced discriminator 2, 6... Pitch search device, 7... Switch, 8... Temporary memory,
9...Pitch search range measuring device, 10...Pitch output terminal.

Claims (1)

[Claims] 1. In a pitch extraction device that obtains a voiced part of an input speech signal and extracts a pitch period from an autocorrelation coefficient sequence obtained for the signal of this voiced part, a first step is performed to obtain a voiced plosive part from an input speech signal. means, second means for removing pitch periods obtained from the signal of the voiced plosive part from pitch periods obtained from the signal of the voiced part; and a pitch period obtained by the second means. and third means for determining a pitch search range when extracting the pitch period based on the period.