Fundamental frequency extraction method and system based on prior surd and sonant knowledge

Abstract

The invention discloses a fundamental frequency extraction method based on prior surd and sonant knowledge, which comprises the following steps of: 1, acquiring a surd and sonant segmentation boundary; 2, calculating the statistic of data to be extracted; 3, combining the statistic acquired in the step 2 with the surd and sonant segmentation boundary which corresponds to the statistic and is acquired in the step 1 for surd and sonant judgment; 4, performing dynamic planning and fundamental frequency track smoothing; and 5, outputting a high-quality fundamental frequency extraction result. In addition, the invention discloses a fundamental frequency extraction system based on the prior surd and sonant knowledge, which comprises a statistic calculation module, a dynamic planning and fundamental frequency smoothing module, a surd and sonant judgment module and a surd and sonant segmentation boundary acquisition module. The surd and sonant segmentation boundary serving as the prior knowledge is input into the surd and sonant judgment module in a fundamental frequency extraction algorithm to assist the surd and sonant judgment, and an ideal surd and sonant judgment result can be acquired by fusing the original automatic surd and sonant judgment with the existing surd and sonant segmentation boundary, so that the accuracy of the surd and sonant judgment is improved.

Description

technical field [0001] The invention belongs to a speech signal processing method, in particular to a fundamental frequency extraction method; in particular to a fundamental frequency extraction method based on prior knowledge of voicing; in addition, the present invention also relates to a method based on prior knowledge of voicing Fundamental frequency extraction system. Background technique [0002] In the current Chinese speech signal processing, it is often encountered that the fundamental frequency parameters need to be extracted. However, many fundamental frequency extraction algorithms now have more or less problems in the judgment of voicing, especially in the speech to be processed. When the pronunciation is not very full and the pronunciation is relatively low, the error rate of voicing judgment is very high, which often leads to a great impact on the subsequent related processing. [0003] In the current speech synthesis research, the parametric speech synthesis...

AI Technical Summary

Problems solved by technology

Most of the voicing and voicing judgment modules of existing fundamental frequency extraction algorithms are based on data statistics. Due to the diversity of human pronunciation, statistical data cannot handle all pronunciation situations well.

Images