Highly expressive speech synthesis method and device

Abstract

The embodiment of the invention discloses a high-expressive speech synthesis method and device. The high-expressive speech synthesis method includes: processing and analyzing the input text to obtain the phoneme sequence corresponding to the input text and the context of the state contained in the phoneme sequence; according to the context of the state, Based on the Viterbi algorithm, select a Gaussian acoustic model from the Gaussian mixture acoustic model containing at least two Gaussian acoustic models corresponding to the state, as the Gaussian acoustic model of the synthesized speech; generate acoustic parameters according to the selected Gaussian acoustic model, and according to Synthesizing the speech with the generated acoustic parameters includes using the vocoder to synthesize the speech with the acoustic parameters or using the acoustic parameters to guide the unit selection of the acoustic segment to generate the speech. The highly expressive speech synthesis method and device provided by the embodiments of the present invention improve the naturalness of synthesized speech.

Description

technical field [0001] Embodiments of the present invention relate to the technical field of text-to-speech conversion, and in particular, to a high-expressive speech synthesis method and device. Background technique [0002] Voice is the most habitual and natural way for man-machine communication. The technology of converting text input into speech output is called text-to-speech (TTS) or speech synthesis technology. It involves many fields such as acoustics, linguistics, digital signal processing and multimedia technology, and is a cutting-edge technology in the field of Chinese information processing. [0003] Since most of the acoustic parameters of the synthesized speech obey the Gaussian distribution, the Gaussian acoustic model is used to generate the acoustic parameters of the synthesized speech, and then the synthesized speech is generated. figure 1 The signal flow of the speech synthesis system based on the Gaussian acoustic model provided by the prior art is sho...

AI Technical Summary

Problems solved by technology

Since the size of the speech synthesis sound library is usually only a few thousand to tens of thousands of sentences, in order to avoid over-training, the leaf node model of the acoustic model decision tree can only represent statistically significant acoustic parameters, and it is difficult to distinguish the differences brought about by the detailed context. difference, thus, resulting in flat, unnatural synthetic speech

Images