High-fidelity intelligent customer service speech synthesis method

Abstract

The invention discloses a high-fidelity intelligent customer service speech synthesis method, comprising: recording speech; establishing an initial corpus; performing forced alignment on the recorded speech; performing speech cutting to obtain a triphone model; using the triphone model to form Target corpus; calculate the likelihood for the trimmed triphone primitives; establish a triphone decision tree; select the appropriate triphone primitive according to the required synthetic text; for the selected triphone primitive, for the original speech waveform Do pitch synchronization labeling; map the pitch curve of the original voice waveform and the target pitch curve to obtain a short-term synthesized signal sequence; arrange the short-time signal sequence and the target pitch cycle synchronously, and overlap and add to obtain a spliced ​​and synthesized voice; The speech is inserted into the pre-recorded speech. The present invention does not require customer service personnel to record a large number of repeated similar voices containing variables, can reduce the cost of voice recording, and can also improve the fluency and naturalness of the entire dialogue process.

Description

technical field [0001] The invention relates to the field of speech synthesis, in particular to a high-fidelity intelligent customer service speech synthesis method. Background technique [0002] The current speech recognition and speech synthesis technology is developing rapidly, but the naturalness of the speech synthesized by the related speech synthesis technology is not very ideal. When it is applied to the customer service voice, the customer will clearly hear the voice of the machine. Most of the customer service voice In fact, it is possible to pre-record the voice, but for variables such as money and time, it is impossible to record a large amount of money, which will consume a lot of money and is not realistic, and there will be obvious differences from the pre-recorded voice, and the entire dialogue process will appear. very strange. Contents of the invention [0003] The technical problem to be solved by the present invention is that, aiming at the above-menti...

AI Technical Summary

Problems solved by technology

[0002] The current speech recognition and speech synthesis technology is developing rapidly, but the naturalness of the speech synthesized by the related speech synthesis technology is not very ideal. When it is applied to the customer service voice, the customer will clearly hear the voice of the machine. Most of the customer service voice In fact, it is possible to pre-record the voice, but for variables such as money and time, it is impossible to record a large amount of money, which will consume a lot of money and is not realistic, and there will be obvious differences from the pre-recorded voice, and the entire dialogue process will appear. very strange

Images