Multi-speaker voice synthesis method, system and device

Abstract

The invention relates to a multi-speaker voice synthesis method and a multi-speaker voice synthesis system. The method comprises the steps: carrying out the voice acoustic statistical feature extraction from a multi-speaker corpus, and obtaining a sentence-level dictionary and a phoneme-level dictionary; extracting text features from the to-be-tested text based on a text analysis method; dynamically combining the text features with a sentence-level dictionary and a phoneme-level dictionary to obtain phoneme-related speaker features; based on an average sub-model and a self-adaptive sub-model,according to the text features and the speaker features, determining speaker voice information; and synthesizing voice through a vocoder based on a neural network according to the voice information ofthe speaker. According to the method, the text features are dynamically combined with the sentence-level dictionary and the phoneme-level dictionary to obtain phoneme-related speaker features, so that the fine description of the speaker features can be improved. Furthermore, voices are synthesized through a vocoder based on a neural network, so that the accuracy of voice synthesis can be effectively improved.

Description

technical field [0001] The present invention relates to the technical field of speech synthesis, in particular to a multi-speaker speech synthesis method, system and device based on phoneme-related speaker features. Background technique [0002] Speech synthesis technology, also known as text-to-speech (Text to Speech) technology, is used to convert text information into voice information. Currently, there are two main methods of speech synthesis: [0003] The first one: a corpus-based speech splicing synthesis method, which is to directly select appropriate primitives from the original recorded corpus to splicing and synthesizing speech. The second method: using parametric speech synthesis method, which is a specific implementation based on statistical acoustic modeling method, which models the acoustic parameters of speech, and reconstructs the trajectory of acoustic parameters through parameter generation algorithms, and finally invokes speech synthesis device to genera...

AI Technical Summary

Problems solved by technology

However, in the traditional method, the features in the speaker recognition task are mostly used, which is not optimal for the speech synthesis task, and lacks a fine description of the speaker's pronunciation characteristics

Images