Speech synthesis method under a small amount of recording samples

Abstract

The invention discloses a speech synthesis method under a small amount of recording samples. The method comprises the following steps of A) obtaining a background speaker model by applying a small amount of training sentences recorded by target recording personnel; B) respectively establishing speaker models for the obtained current speaker sentence and the originally recorded complete speaker sentence through a self-adaptive algorithm; and C) realizing the speech synthesis by adjusting the mean and variance of the speaker models, and synthesizing a small amount of recorded audios into a complete recording sentence. The speech synthesis method under a small amount of recording samples disclosed by the invention has the advantages that an operation of repeatedly recording the same recordingtext by multiple customer service staff is avoided, the speed recording cost is reduced, and the fluency and the naturalness of the whole conversation process effect can be guaranteed.

Description

technical field [0001] The invention relates to the field of speech synthesis, in particular to a speech synthesis method under the condition of a small number of recording samples. Background technique [0002] The speech-related fields of Dang, especially speech synthesis, are developing extremely rapidly, but the preparation process of the corpus is very difficult. During the recording process, when different recording personnel record the same content, a lot of time, energy and financial resources may be wasted due to too much recording content, and it is very unnecessary to record the same recording text multiple times. The naturalness of the related speech synthesis technology for a small amount of speech synthesis is not ideal, and at the same time there will be obvious differences with the actual recorded speech, which cannot achieve the desired ideal effect. Contents of the invention [0003] The technical problem to be solved by the present invention is that, ai...

AI Technical Summary

Problems solved by technology

[0002] Speech-related fields, especially speech synthesis, are developing extremely rapidly, but the process of preparing corpora is very difficult

During the recording process, when different recording personnel record the same content, due to the excessive recording content, a lot of time, energy and financial resources may be wasted, and it is very unnecessary to record the same recording text multiple times

The naturalness of the related speech synthesis technology for a small amount of speech synthesis is very unsatisfactory, and there will be obvious differences from the actual recorded speech, which cannot achieve the desired ideal effect

Images