Research On Command And Control Oriented Speech Recognition System Technology

The artificial intelligence industry is booming.As the most natural and convenient way of communication for human beings,the interactive method using voice input has become an important interface for future human-computer interaction.With the accompanying development,voice-based interactive input and control has become an extremely important means of interaction.Different from the traditional civilian scene and the military command and control scene,the harsh battlefield use environment places more stringent requirements on the voice interaction system.Faced with these special use scenarios,how to improve and improve the speech recognition system in a targeted manner has become an urgent problem.The research aimed at individual soldiers ’companion equipment.Firstly,a speech recognition service architecture suitable for individual soldiers’ accusation system was combined with the concepts of local speech recognition,cloud speech recognition and edge computing.This architecture improves the intelligent voice interaction experience as much as possible on the premise of adapting to the demanding application scenarios.And when the model is deployed,the training and recognition of the speech recognition model is more implemented on the terminal device,thereby reducing the network transmission burden and releasing the computing power in the cloud.The hardware technology、software algorithm of the edge terminal speech recognition system in the previous architecture are studied.Based on this,a pruning neural network update training algorithm suitable for edge terminal deployment is proposed.The experimental results show that this method is not only conducive to the promotion of speech recognition pruning model deployment at edge terminals,but also provides a feasible model update Program.Research on related leading-edge algorithms.An improved extended hybrid end-to-end Chinese speech recognition model based on CTC and attention mechanism is proposed,which improves the training rate to a certain extent and makes the speech recognition system perform better in noisy environments.At the end of this paper,we make a preliminary engineering exploration on the corresponding prototype system.