Method for labeling audio by using deep learning model

Abstract

The invention discloses a method for labeling an audio by using a deep learning model. The method comprises the following steps: A, acquiring the audio and performing voice preprocessing on the acquired audio; b, inputting the audio data subjected to voice preprocessing into a deep learning model for voice recognition and voice annotation, and labeling the audio according to the voice annotation,wherein the deep learning model comprises a deep neural network and a long-short-term memory unit; and C, performing manual proofreading on the label output by the deep learning model. According to the method disclosed by the invention, the tedious work of manual listening, manual labeling and manual proofreading is converted into the work of only needing manual proofreading, and other operationsare automatically carried out by the system model, so that the manpower and time cost can be greatly saved, and the effectiveness is guaranteed.

Description

technical field [0001] The invention relates to the technical field of speech recognition, in particular to a method for marking audio by using a deep learning model. Background technique [0002] In the field of deep learning speech recognition, sufficient raw corpus data is required before training, and keywords and invalid speech in the corpus data are marked. Labeling keywords is an important preprocessing process in speech signal processing systems such as speech recognition and speech enhancement. Due to the huge amount of corpus data, if it is based on traditional voice tagging methods, the work of tagging keywords is cumbersome and consumes a lot of manpower and time. At the same time, with the rapid development of artificial intelligence, new opportunities and challenges have been brought to speech recognition, and there is an urgent need for a speech annotation method that can reduce manpower and time costs. [0003] At present, most of the automatic speech recog...

AI Technical Summary

Problems solved by technology

[0003] At present, most of the automatic speech recognition technologies at home and abroad rely on a large number of data resources, and these data resources need to mark the speech through traditional means. In the patent CN201811011859.7, a method for low-resource Tujia The end-to-end speech recognition method of Chinese language, this method is to improve the recognition rate through convolutional neural network and BiLSTM, the main purpose of this method is to improve the result of speech recognition, that is, to improve the recognition rate, for speech recognition, most of them are based on Pure speech, but not very good for noisy speech data

Images