Speech synthesis model training method, speech synthesis method, speech synthesis device and medium

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a speech synthesis model training method. The method comprises the following steps: obtaining a training text, and carrying out phoneme conversion on the training text by using a preset speech synthesis model to obtain a phoneme sequence, wherein speech synthesis model comprises an encoder, a decoder, a residual network and a WaveGlow vocoder; sequentially processing the phoneme sequence by using the encoder, the decoder and the residual network to obtain a target Mel spectrum; performing parallel audio conversion on the target Mel spectrum by using the WaveGlow vocoder to obtain a target audio; and performing loss calculation on the target audio to obtain a training loss value, and adjusting parameters of a speech synthesis model according to the loss value to obtain a target speech synthesis model. In addition, the invention also relates to a speech synthesis method and device, equipment and a storage medium. According to the invention, the accuracy of the speech synthesis model can be improved, and the speech synthesis training speed of the speech synthesis model is accelerated.

Description

technical field [0001] The present application relates to the technical field of artificial intelligence, and in particular to a speech synthesis model training method, speech synthesis method, device, electronic equipment and storage medium. Background technique [0002] Speech synthesis refers to the technology of converting any input text into corresponding speech, which is an important research branch in the field of natural speech processing. [0003] At present, the common speech synthesis model is the Tacotron2 model. The Tacotron2 model uses character embedding as the input model during training. Therefore, a large training corpus is needed to ensure that the trained Tacotron 2 model will not have pronunciation errors. However, in the actual application process, Usually, a large number of training corpora cannot be obtained, resulting in low speech synthesis accuracy, and even obtaining a large number of training corpora will slow down the speed of speech synthesis. ...

AI Technical Summary

Problems solved by technology

[0003] At present, the common speech synthesis model is the Tacotron2 model. The Tacotron2 model uses character embedding as the input model during training. Therefore, a large training corpus is needed to ensure that the trained Tacotron 2 model will not have pronunciation errors. However, in the actual application process, It is usually impossible to obtain a large number of training corpora, which leads to low speech synthesis accuracy, and even obtaining a large number of training corpora will slow down the speed of speech synthesis

In addition, the Tacotron2 model uses WaveNet as the vocoder. Since the WaveNet vocoder uses a sample-by-sample speech waveform generation mechanism, the synthetic speech speed is very slow and the accuracy is not high.

Images