Chinese rhythm hierarchy prediction method and system based on self-attention

Abstract

The invention discloses a Chinese rhythm hierarchy prediction method based on self-attention. The method comprises the steps of learning a large number of unlabeled texts to obtain word vectors of single words, converting a to-be-predicted text into a word vector sequence by utilizing the word vectors, inputting the word vector sequence into a trained rhythm level prediction model, and outputtingword positions and rhythm levels of the text. According to the method, Chinese rhythm hierarchy prediction is carried out by using a rhythm hierarchy prediction model, characteristics of character granularity are used as input while prediction performance is guaranteed, dependence on a word segmentation system and possible negative effects are avoided, the model directly models the relation between any two characters in a text through a self-attention mechanism, and parallel calculation can be achieved. Pre-training is carried out by using additional data to improve the model performance, so that each rhythm level of the to-be-processed text can be accurately predicted at the same time, and wrong transmission is avoided.

Description

technical field [0001] The invention relates to the technical field of speech synthesis, in particular to a method and system for predicting Chinese prosody levels based on self-attention. Background technique [0002] In the speech synthesis system, predicting the prosodic hierarchical structure according to the input text to be synthesized has always been a crucial step, and the prediction result will be used as a part of the linguistic features to model the acoustic features and duration. Therefore, the accuracy of prosodic level prediction largely determines the naturalness of synthesized speech, and it is of great significance to realize accurate prosodic level prediction. [0003] The current mainstream method is to use the bidirectional long-short-term memory network BLSTM to model different prosodic levels with word vectors as input, that is, to train a model for prosodic words, prosodic phrases, and intonation phrases, and use the low-level prediction results A lev...

AI Technical Summary

Problems solved by technology

[0004] However, the above method has the following problems: 1) LSTM, as a RNN structure, needs to use the output value of the previous moment every time it predicts the output value at the current moment. This sequential calculation hinders its parallelization and makes any two The distance between words is O(n); 2) Training and predicting the prosody prediction model at the word granularity means that the input text must be segmented first, and the result of word segmentation will directly affect the performance of prosody level prediction

In addition, the number of Chinese words is huge, and storing these word vectors will take up a lot of storage space and computing resources, which is obviously not practical for offline speech synthesis; 3) step-by-step prosody prediction will make wrong results continue to be transmitted , leading to subsequent prediction errors

[0005] Predicting the prosodic level of text is an essential step in the speech synthesis system, but the current mainstream method uses word-level features to rely on the performance of the word segmentation system, and prosodic prediction level by level will cause continuous transmission of wrong results

Images