Speech pitch estimation method and device

Abstract

The invention relates to a voice pitch period estimation method and device. The device comprises a signal preprocessing unit, a normalized autocorrelation function computing element and a pitch period postprocessing unit. The method includes the steps of firstly, conducting preprocessing including direct current component removal, perception weighting and under-signal sampling on voice signals; secondly, computing normalized autocorrelation function values of the processed voice signals; thirdly, determining the maximum of the normalized autocorrelation function values in the pitch period searching range, and determining a pitch period candidate value corresponding to the maximum to be a pitch period estimation value of the voice signals. According to the voice pitch period estimation method and device, frequency doubling errors and frequency halving errors in the pitch period estimation are well overcome, the noise resistance performance of the pitch period estimation method is improved, meanwhile, the algorithm complexity of an algorithm is lowered, and the corresponding digital audio / speech coding efficiency is improved. The voice pitch period estimation method and device can be applied to pitch searching of various voice coding and decoding algorithms and have a wide application range.

Description

technical field [0001] The present invention relates to speech coding technology, more specifically, to a speech pitch period estimation method and device. Background technique [0002] The pitch period refers to the period in which the vocal cords vibrate when a person makes a sound. Pitch period is an important issue in speech coding, and its accuracy will directly affect the coding quality and efficiency of the speech coder. Accurate pitch periodicity analysis can effectively remove redundancy in the speech coding process, reduce the number of coding bits, and realize low bit rate high-quality speech coding. However, due to the particularity of speech, the accurate search of the pitch period will face the following difficulties: [0003] (1) The change of the speech signal is very complicated, the glottal excitation waveform is not a complete periodic pulse train, and the period of the speech waveform is time-varying. [0004] (2) At the beginning and end of speech, th...

AI Technical Summary

Problems solved by technology

[0003] (1) The change of the speech signal is very complicated, the glottal excitation waveform is not a complete periodic pulse train, and the period of the speech waveform is time-varying

[0004] (2) At the beginning and end of the speech, there is no periodicity like vocal fold vibration. It is difficult to determine whether some transitional sounds such as unvoiced and voiced sounds belong to periodic or non-periodic signals, so it is impossible to estimate the pitch period

[0005] (3) To remove the influence of the vocal tract from the speech signal, it is difficult to directly extract information related to the vibration of the vocal cords

[0006] (4) The difficulty of defining the exact start and end of each pitch period in the voiced segment limits the reliable measurement of pitch, not only because the speech signal itself is quasi-periodic (that is, the pitch changes), but also because the waveform is easy to Affected by formants, noise, etc.

[0008] (6) The large range of pitch cycle changes also brings certain difficulties to accurate pitch detection

However, the ACF method in the general time domain is prone to "multiple frequency" and "half frequency" errors, and the AMDF method cannot effectively track rapid changes in speech frequency

The frequency domain method generally adopts the cepstrum method. Due to the introduction of logarithmic operation, the calculation amount is greatly increased, and it is easily affected by noise.

Images