Speech synthesis related method, speech stream speech change model training method and related device

Abstract

The invention discloses a speech synthesis related method, a speech stream speech change model training method and a related device, and the speech synthesis method comprises the steps: carrying out the pinyin labeling of a to-be-processed text, and obtaining the pinyin labeling information of the to-be-processed text; inputting the to-be-processed text and the pinyin annotation information of the to-be-processed text into a speech stream sound change model to obtain first phonetic symbol stream data of the to-be-processed text; and performing speech synthesis on the to-be-processed text based on the first phonetic symbol stream data. Through the above steps, the speech synthesis method can improve the naturalness of the synthesized speech.

Description

technical field [0001] The present invention relates to the field of speech synthesis, in particular to a speech synthesis related method, a training method of a speech flow-to-speech model and a related device. Background technique [0002] TTS (text to speech, also known as speech synthesis, text-to-speech conversion) refers to the process by which a machine converts language from a text carrier to a sound carrier, and is a key module in systems such as man-machine dialogue and intelligent broadcast. With the maturity of related technologies, the competition of speech synthesis products of major manufacturers gradually focuses on the naturalness of the synthesized speech. [0003] However, there are strong mechanical sounds in the current synthesized speech, which seriously affect the user's experience, and the mechanical sounds are mainly caused by two types of reasons. Consecutive readings should be read consecutively; the second is that the pronunciation of the languag...

AI Technical Summary

Problems solved by technology

[0003] However, there are strong mechanical sounds in the current synthetic speech, which seriously affects the user's experience. The mechanical sounds are mainly caused by two types of reasons. What should be read together should be read together; the second is the wrong pronunciation of language flow, such as: taking Mandarin Chinese as an example, the first upper tone of continuous reading (for example: "tiger") should be pronounced as Yangping tone (second tone) , but still read the upper tone (three tones)

[0004] Existing speech synthesis methods are difficult to reduce the mechanical sound of synthesized speech

Images