Layer-by-layer channel selection method for voice recognition of self-organizing microphone

Abstract

The invention discloses a layer-by-layer channel selection method for voice recognition of a self-organizing microphone, and the method is based on a conformer voice recognition framework, and the method comprises: (1), an encoder-decoder framework is adopted, an encoder is based on a Conformer framework, a decoder is based on a Transformer framework, and a multi-head attention mechanism is introduced into an encoder-decoder module; (2) for a single-channel voice recognition system, clean voice is adopted for independent training; and (3) for a multi-channel voice recognition system, the voice of each channel is encoded and then the same decoder is shared, a multi-layer flow attention mechanism is trained, and the channels are screened layer by layer. Under a large-scale self-organizing microphone array, compared with other flow attention-based methods, the method provided by the invention is higher in speech recognition accuracy and lower in calculation complexity.

Description

technical field [0001] The invention belongs to the technical field of voice recognition, and in particular relates to a layer-by-layer channel selection method for voice recognition. Background technique [0002] Long-distance speech recognition is an extremely challenging problem. Multi-channel speech recognition based on microphone array is an important way to improve performance. However, when the distance between the speaker and the microphone array increases, the quality of speech will drop sharply, resulting in a physical upper bound for the performance of Automatic Speech Recognition (ASR) no matter how many channels are added to the array. Self-organizing microphone array is a method to solve the above problems, which includes a series of microphone nodes randomly scattered in the sound field, and the microphone node can be a microphone or a microphone array. Using channel weight assignment and channel selection, the microphones around the speaker can be automatic...

AI Technical Summary

Problems solved by technology

However, the above methods only consider the channel weight assignment of a small number of self-organizing nodes (no more than 10 microphone nodes), and do not explore the issue of channel selection

When the sound field environment becomes larger and more complex, and there are more self-organizing nodes, on the one hand, because some channels are greatly affected by noise, some channels that do not help the performance need to be discarded; on the other hand, due to the increase in the number of channels, the network The amount of calculation increases, so it is necessary to explore channel selection methods that reduce computational complexity

Images