A Speech Synthesis Method Based on Generative Adversarial Networks

Abstract

A method for speech synthesis based on generative confrontation network, comprising the following steps: S1. preparing training samples, including real audio data, extracting Mel spectral features; S2. setting initialized generator parameter groups and discriminator parameter groups; S3. Input the Mel spectrum feature to the generator to obtain the output synthetic audio; S4. Input the real audio data in S1 and the output synthetic audio corresponding to S3 to the multidimensional discriminator and the pooling discriminator; S4. The output result of the discriminator Input to the loss function formula to calculate the generator loss function and discriminator loss function respectively; S5. Update the generator and discriminator; S6. Use the updated generator and discriminator to repeat until reaching the set maximum number of updates M; S7. After each update, return to step S3; S8. Use the generator for speech synthesis. The generator of the present invention adopts one-dimensional convolution operation, the model parameters are small, and the speed is fast.

Description

Technical field [0001] The present invention belongs to the field of manual intelligence speech synthesis, and in particular, the present invention relates to a speech synthesis method based on a generating counterfeit network. Background technique [0002] The voice is the most direct and fast communication method, which plays a very important role in the field of manual intelligence, has been widely used in robots, automotive, synthetic anchors and other fields. With the wide application of artificial intelligence products, the naturalness, clarity, understandability of speech synthesis is also higher and higher. Deep learning makes voice synthesis techniques grow rapidly. [0003] The commonly used speech synthetic scheme now is mainly divided into two phases: predicts its acoustic characteristics based on text information, such as MEL-Spectrograms; It is predicted by acoustic character predicting its original audio waveform, that is, a codon model learning. The first stage is...

AI Technical Summary

Problems solved by technology

WaveNet is an autoregressive convolutional neural network. As the first batch of deep learning algorithms used for speech synthesis, it greatly improves the quality of speech synthesis. However, its model structure determines that the speed is very slow and it is difficult to apply it to actual products.

In recent years, the research on the speech synthesis vocoder has mainly focused on improving the calculation speed and reducing the model parameters, and there is no significant improvement in the synthesis speed

Images