Design Of Speech Recognition System Based On Microphone Array

In recent years,human-machine voice interaction with "intelligent voice recognition" as the core is under rapid development.Far-field voice recognition technology based on microphone arrays is widely used.In practical application scenarios,voice signals are inevitably mixed with various noises,which seriously affects the accuracy of speech recognition.An excellent speech recognition system not only needs to perform well in its own model,but is also closely related to the microphone array used to collect the signal and the quality of the input voice signal.This paper designs a microphone array module for speech recognition signal acquisition;using this microphone array as a radio device,a set of speech enhancement algorithms is proposed;Integrated microphone array module,voice enhancement module,and voice/text conversion module to complete the voice recognition system.The test results show that for the original noisy speech,the speech recognition system completed in this article has significantly improved accuracy and stability than the speech recognition system using traditional algorithms.The research carried out in this article is as follows:1.Microphone array module design.Carried out the microphone array structure design;completed the microphone array hardware circuit design,including microphone selection,multiple selection circuit design,signal amplification circuit design,analog-to-digital conversion circuit design and embedded MCU design.2.Speech enhancement algorithm research and testing.The algorithm includes four parts: automatic gain control,adaptive noise reduction,voice activity detection,sound source localization and beam forming.Among them,adaptive noise reduction and voice activity detection play a key role in the accuracy of subsequent voice recognition.After the original noisy speech is processed by the improved spectral subtraction noise reduction algorithm proposed in this article,the energy-entropy ratio speech endpoint detection algorithm based on the dual-threshold method and the sound source localization and beamforming algorithm are processed,the pure human voice can be separated,which is convenient for subsequent speech Recognition.3.Integrated microphone array module,voice enhancement module,and voice/text conversion module to complete the voice recognition system.Based on the MFC library,complete the system application interface design.Test the sound source localization algorithm based on the microphone array and the voice/text conversion module.The test result proves that the accuracy and stability of speech recognition are significantly improved after the speech signal is enhanced.