부산대학교 도서관

In order to understand the application of Russian phonetic information processing and conversion, an application research based on greedy algorithm in Russian phonetic information processing and conversion is put forward. In this paper, firstly, the word-to-sound conversion is regarded as a statistical classification problem, and by determining the Russian characteristics and constraints, the maximum criterion and index, the maximum multi-tone word disambiguation model is constructed. The Russian dictionary is expressed as Markov model, and the language model combining rules and statistics is used to evaluate all conversion paths, and a language analysis tree is established. According to various assumptions and evaluation functions, the optimal conversion path is obtained. Secondly, the greedy algorithm is used to add regularization factors, and the speech signal is set to an appropriate length to complete the rapid conversion of words and sounds. Finally, 500 Russian sentences are randomly selected from TIMIT database for conversion test. The experimental results show that the Russian speech signal output by the proposed method is relatively stable, and the signal loss and conversion error rate are reduced in the word-to-sound conversion.