An adaptive method and system for speech recognition based on a cached language model

Abstract

The invention discloses a voice recognition self-adaptive method and system based on a cached language model, belonging to the field of voice recognition. In the present invention, by receiving the continuous voice signal input by the user, the continuous voice signal is divided into multiple short voices based on the voice activity detection technology VAD, and the short voices are sequentially recognized based on the general language model, and a corresponding identification is generated for each short voice. As a result, the associated vocabulary is obtained based on keyword search, and the associated vocabulary is processed through the cache model to obtain a language model that adapts to local changes in the distribution of historically recognized texts. Based on the modified language model, the subsequent short speech continues to be recognized. After partial modification, the language model has a better similarity with the historical recognition content, which improves the recognition accuracy of continuous long speech. In addition, users can actively correct misrecognized low-frequency words during the recognition process to improve the subsequent recognition accuracy of low-frequency words.

Description

technical field [0001] The invention relates to the field of speech recognition, in particular to an adaptive method and system for speech recognition based on a cached language model. Background technique [0002] After decades of development, speech recognition has become a relatively mature technology. In practical applications, Siri, Cortana, etc. have high recognition accuracy under ideal conditions. [0003] The performance of a speech recognition system largely depends on the similarity between the language model (LM) used and the task to be processed. This similarity is especially important in cases where the statistical properties of language change over time, such as in application scenarios involving spontaneous and multi-domain speech. Topic Identification (TI) based on information retrieval is a key technology. The topic under discussion can be obtained through semantic analysis of historical recognition results, so as to adjust the language model and realize d...

AI Technical Summary

Problems solved by technology

[0006] In order to solve the problem that the existing speech recognition system based on single-sentence tasks cannot adapt to domain information to recognize low-frequency words, resulting in low recognition accuracy of low-frequency words or too complicated recognition system, the present invention proposes a speech recognition system based on a cached language model. Identify adaptive methods and systems

Images