Speech synthesis model training method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention relates to the technical field of natural language processing, and discloses a speech synthesis model training method and device, electronic equipment and a storage medium, the model comprises a generator and a discriminator, and the training method comprises the following steps: obtaining a plurality of first audio data marked with text information to generate a training sample set; inputting the text information into the generator, and obtaining second audio data output by the generator; the first audio data and the second audio data are input into the discriminator, a discrimination result output by the discriminator is obtained, and the discrimination result is used for representing the similarity degree between the first audio data and the second audio data; according to the judgment result and the preset loss function, iterative training is performed on the speech synthesis model, an intermediate conversion link is not included, accumulative errors are eliminated, the speech synthesis effect of the model is improved, the model can be trained to be converged only through a small number of training samples, and the training cost is reduced.

Description

technical field [0001] The embodiments of the present application relate to the technical field of natural language processing, and in particular to a training method, device, electronic device and storage medium of a speech synthesis model. Background technique [0002] Speech synthesis technology can convert text information into corresponding speech information. Traditional speech synthesis technology is generally divided into speech synthesis technology based on statistical parameter modeling (also known as parametric synthesis), speech synthesis technology based on unit selection and waveform splicing (also known as parametric synthesis). Splicing synthesis), and speech synthesis technology based on neural network model, etc., among them, the speech synthesis technology based on neural network model first converts text information into specific acoustic features, such as Mel spectrum, and then uses vocoder to convert Mel Acoustic features such as spectrum are converted ...

AI Technical Summary

Problems solved by technology

[0003] However, the inventors of the present application have discovered that the speech synthesis technology that converts text information into specific acoustic features first, and then converts acoustic features into audio waveform data is a two-stage technology that converts text information into specific acoustic features. The acoustic features are realized by the acoustic model, and the conversion of the acoustic features into audio waveform data is realized by the vocoder. The acoustic model and the vocoder need to be trained separately, and the trained acoustic model and the vocoder are combined in series to form When the speech synthesis system is used, it will bring cumulative errors, resulting in low quality of the generated speech information. If the acoustic model and the vocoder are spliced ​​together before training, a large number of training samples are needed to make the model converge, which greatly increases the training cost and training time

Images