Chinese Prosodic Hierarchical Structure Prediction System

Abstract

The invention discloses a Chinese prosodic hierarchy prediction system. The system comprises a text analysis module, a text feature parameterization module, a character-term vector joint training module, a term vector generation module, a first single classifier training module, a second single classifier training module, a feature importance ranking module and a model fusion module; the text analysis module outputs text data analyzed; the text feature parameterization module outputs parameterized text features; the character-term vector joint training module receives the analyzed text data generated by the text analysis module, and outputs a term vector representation model using character vectors for enhancement for the text; the term vector generation module utilizes the term vector representation model using the character vectors for enhancement to output term vectors, enhanced with the character vectors, of the analyzed text data; the first single classifier training module outputs a first mapping module; the second single classifier training module outputs a second mapping model; the feature importance ranking module outputs text parameter features with preset classification performance; the model fusion module outputs results of prosodic hierarchy prediction. Through the system, accuracy in prosodic hierarchy prediction is improved.

Description

technical field [0001] The embodiments of the present invention relate to the technical field of total human-computer interaction speech synthesis, and in particular to a Chinese prosodic hierarchical structure prediction system. Background technique [0002] Accurate prosodic hierarchy description and prediction of prosodic hierarchy structure from text information has always been a crucial step in speech synthesis, and is an important part of improving the naturalness and expressiveness of synthesized speech and building a harmonious human-computer interaction technology. The prosodic structure model can describe the cadence and priority of the speech, thereby improving the expressiveness and naturalness of the synthesized speech. Prosodic structure modeling and prediction is of great significance to the development of speech synthesis and human-computer interaction. [0003] Although there have been a lot of research work in this field, there are still many problems in p...

AI Technical Summary

Problems solved by technology

[0005] 1. The description of text features is not precise enough

Most of them only describe the text from some traditional features such as part of speech and word length; while few consider the description of text features by word vector features, and do not consider that a word may have different meanings in different contexts. The meaning and the possible influence of the words inside the word on the word vector

[0006] 2. The prediction accuracy of a single model generally cannot reach an ideal state, so the naturalness of speech synthesis is also greatly damaged, which in turn affects people's sense of hearing

Images