Speech recognition method, device and electronic equipment

Abstract

An embodiment of the present disclosure provides a speech recognition method, device, and electronic equipment, which belong to the field of data processing technology. The method includes: using pre-set training samples to randomly initialize and train the LSTM network model; based on the set and CTC loss function The related first training parameter and the second training parameter related to KL divergence utilize the training result of the LSTM network to form a loss function for training the BLSTM network; in the process of training the BLSTM network, gradually increasing While increasing the value of the first training parameter, gradually reduce the value of the second training parameter; when the performance index output by the BLSTM network meets the preset requirements, stop the training of the BLSTM network, so that Use the BLSTM network to perform real-time text prediction on the input sound file. Through the processing solution disclosed in the present disclosure, the accuracy of the time stamp predicted by the acoustic network model can be improved.

Description

technical field [0001] The present disclosure relates to the technical field of data processing, and in particular to an acoustic network model training method, device and electronic equipment. Background technique [0002] Speech processing (Speech processing), also known as voice signal processing, human voice processing, its purpose is to make the desired signal, further do voice recognition, apply to the mobile phone interface and even ordinary life, so that people and computers can communicate. [0003] In the process of speech processing, the audio-like sound signal received by a microphone or other devices can be used to process the data through an analog-to-digital conversion device, and finally output through a digital-to-analog conversion device. Therefore, the processing is for digital signals, and the speech signal is a discrete time signal. The signal processing flow is as follows: Collect and sample signals: use a microphone or various radio devices to collect...

AI Technical Summary

Problems solved by technology

The signal processing flow is as follows: Collect and sample signals: use a microphone or various radio devices to collect analog voice signals, and then use an ADC device (such as an analog-to-digital conversion card) to convert the analog signal into a digital signal, and then according to the Nyquist theory For sampling, if it does not conform to the theory, it will cause signal distortion

Images