Neural network training method and device and audio processing method and device

Abstract

The invention relates to a neural network training method and device and an audio processing method and device, and the method comprises the steps: inputting a training audio into a coding sub-network for coding, obtaining a first coding result, inputting a text label corresponding to the training audio into a prediction sub-network for prediction, and obtaining a first prediction result; combining the first coding result with the first prediction result to obtain a first combination result; according to the first joint result, respectively cutting the first coding result and the first prediction result to obtain a corresponding second coding result and a corresponding second prediction result; and inputting the second coding result and the second prediction result into a joint sub-network for joint processing to obtain a second joint result, and adjusting network parameters of the coding sub-network, the prediction sub-network and the second joint sub-network according to the second joint result.

Description

technical field [0001] The present disclosure relates to the technical field of audio processing, and in particular, to a neural network training method and device, and an audio processing method and device. Background technique [0002] In recent years, audio processing technologies such as audio recognition have gradually developed, and their accuracy has become higher and higher, and has played an important role in many fields. At present, the audio processing field has temporal connection classification (Connectionist TemporalClassification, CTC) technology, attention-based model (Attention-based model) technology and recurrent neural network converter (RNN Transducer, RNN-T) technology, among which the recurrent neural network Converter technology works best in practice. However, in the cyclic neural network converter technology in the related art, the operation during neural network training is relatively complex, and the memory occupies a lot, which makes it impossib...

AI Technical Summary

Problems solved by technology

However, in the cyclic neural network converter technology in the related art, the calculation of the neural network training is relatively complicated, and the memory usage is large, which makes it impossible to achieve fast calculation, which greatly reduces the practicability of the technology.

Images