Acoustic model optimization method and device

Abstract

The invention provides an acoustic model optimization method and device, and belongs to the technical field of voice transfer. The method comprises the following steps: acquiring audio data of a preset length time period before the current moment; extracting effective voice data in the audio data, determining the effective voice duration corresponding to the effective voice data and effectively transferring the text; according to the threshold value range of the effective voice duration, determining a parameter updating rule, training the acoustic model through the effective voice data and theeffective transferring text, so as to update the parameters specified in the acoustic model by the parameter updating rule. Due to the fact that a large amount of relevant audio data is collected inadvance, the collection difficulty corresponding to the mode that the audio data is acquired in real time is small, and the workload is small. In addition, the audio data of the speaking user can be acquired in real time in the actual use scene, so that the obtained audio data and the actual use scene are matched with each other, and are matched with the speaking user, and the optimization effectis better when the acoustic model is optimized.

Description

technical field [0001] Embodiments of the present invention relate to the technical field of speech transcription, and more specifically, relate to an acoustic model optimization method and device. Background technique [0002] With the continuous development of speech recognition technology, speech transcription is also being more and more widely used. Speech transcription mainly refers to the conversion of voice into text, which is often used in meeting minutes, customer phone recording analysis, and court trial data recording scenarios. Speech transcription can greatly reduce the manual workload in these scenarios and improve work efficiency. At present, when transcribing speech, it is mainly realized by acoustic model and language model. Among them, the acoustic model is used to recognize the acoustic features of speech as phonemes or words, and the language model is used to decode phonemes or words into a complete transcribed text. Acoustic models usually need to be ...

AI Technical Summary

Problems solved by technology

[0003] Due to the need to collect a large number of relevant audio data of scenes or speakers, and the collection of these relevant audio data is difficult and the collection workload is relatively large, so it is time-consuming

In addition, because the relevant audio data collected in advance may not match the actual usage scenarios, resulting in poor recognition of the acoustic model

Images