Word acoustic feature system, training method and system for word acoustic feature system

Abstract

Embodiments of the present invention provide a method for training a word acoustic feature system. The method includes: splicing word acoustic features output by a word acoustic feature system with a phoneme feature sequence output by a phoneme encoder to obtain a phoneme feature sequence with word acoustic features, and splicing with actual prosody features to obtain a phoneme feature sequence with prosody and word acoustics The phoneme feature sequence of the feature; adjust the coding length, add the pitch and energy features, and decode to obtain the predicted mel spectrum; train the word acoustic feature system based on the actual mel spectrum and the predicted mel spectrum. Embodiments of the present invention also provide a word acoustic feature system and a training system for the word acoustic feature system. The embodiment of the present invention uses the trained word acoustic feature system to obtain word acoustic features that not only have word meanings, but also pronunciation, and through continuous training of the word acoustic feature system, the word acoustic features are more accurate, thereby further improving speech synthesis. The quality of speech synthesis.

Description

technical field [0001] The invention relates to the field of intelligent speech, in particular to a word acoustic feature system, a training method and system for the word acoustic feature system. Background technique [0002] End-to-end text-to-speech synthesis models with sequence-to-sequence architectures have achieved great success in generating natural speech. The word features are characterized by aligning and splicing the word vector encoder with the phoneme feature sequence (the output of the phoneme encoder) through text analysis or extracting the word vector representation from the pre-training model. Ways to obtain these feature vectors include: [0003] Obtain word features by statistical methods, such as word frequency, etc., and then use text analysis to generate word feature vectors; [0004] Extract the encoder output as word vectors from common machine learning tasks (such as translation tasks); [0005] Use the BERT coding layer to extract word vectors; ...

AI Technical Summary

Problems solved by technology

[0009] In order to at least solve the problem that the existing models in the existing methods only focus on the meaning of the word and ignore the pronunciation of the word, making the feature vector less effective in improving the quality of speech synthesis

Images