Speech segment splicing system and method for speech synthesis

Abstract

The present invention relates to a system and method for splicing speech segments for speech synthesis. First, two speech segments to be spliced ​​are extracted from a speech bank as a first speech segment and a second speech segment, and the first speech segment and the second speech segment are Select the best sampling point in the second speech segment; then, carry out first-order smoothing to the best sampling point, generate speech splicing point; First-order smoothing method is: calculate the slope ka, kb at the best sampling point U1, U2 place, and The numerical difference value deltaU of the best sampling points U1 and U2; predict according to the slope ka, kb and the difference value deltaU, and generate a speech splicing point. Finally, a speech splicing point is inserted between the first speech segment and the second speech segment to generate a third speech segment. The present invention solves the problem of voice spectrum hopping that occurs in direct splicing in the prior art, and the problem of excessive calculation through the autocorrelation search and accumulation smoothing method, and obtains a good continuous frequency spectrum at the splicing place through the first-order smoothing method. and enhance the user's hearing experience.

Description

technical field [0001] The invention relates to the field of speech synthesis, in particular to a speech segment splicing system and method for speech synthesis. Background technique [0002] There are two existing speech synthesis methods based on speech feature parameters and based on waveform splicing. Compared with the parameter-based method, the speech synthesis based on waveform splicing can obtain higher-quality synthesized speech, and the sound sounds more natural, which is closer to the timbre of the original speaker. Therefore, the current mainstream online speech synthesis focuses on the speech synthesis scheme based on waveform splicing. [0003] The principle of the speech synthesis method based on waveform splicing is as follows: first select the appropriate speech unit from the pre-recorded and marked speech library as the speech segment to be spliced, and then obtain the final synthesized speech by splicing the speech segments. Using this splicing method, i...

AI Technical Summary

Problems solved by technology

Using this splicing method, if the spliced ​​fragments are not handled well at the connection, there will be jumps in the frequency spectrum, which will lead to unnatural hearing experience for users

Therefore, a key technical issue is: what kind of splicing method is used to make the spliced ​​speech clips output smoothly?

[0004] At present, the existing splicing method adopts the method of aligning the speech segments first and then accumulating and smoothing them. The smoothing effect of the speech segments output by this splicing method is average, and there is a problem of jumping between the frequency spectrums of the speech segments.

In addition, in some cases, this stitching method has the problem of not finding a smooth alignment point

Images