Research And Realization Of Isolated-word Speech Recognition System Based On DSP

This article is based on TMS320VC5502chip，designs of a non-specific embeddedspeech recognition system．The system uses the ping-pong type data processingmethods to guarantee that the acquisition and recognition of speech signal in real time；the use of online activity detection algorithm can effectively detect speaking voice inthe audio stream；you can overcome the flaws of the traditional method of DTW，improve recognition accuracy rate of the system，with the acoustic models based onhidden Markov model．The implementation of speech recognition system is mainly made up of DSPhardware devices and the corresponding software algorithm．High-Speed DSP chip usesTI company TMS320VC5502as the system’s hardware platform，speech signal is inputthrough a microphone，and then collected and processed by TLV320AIC23B audiocodec chip．The DMA manager is responsible for sending the speech signal to the datastore．Through Endpoint Detection and feature extraction，the voice signal is convertedto the coefficient of mfcc features，and finally you will get the final result of recognitionby Viterbi algorithm．The system is mainly composed of pretreatment of speech signal、Endpoint Detection、feature extraction、the similarity measure and so on．Endpoint Detection Method Based on energy can increase the robustness ofdetection，reduce the decline in the detection of risks caused by noise，compared withtraditional method based on energy and zero-crossing rate threshold．The means ofusing the acoustic model based on hidden Markov as the identification model，has goodscalability and identify performance．Speech is the primary means of human communication．Fields of embeddedsystems is an important application of speech recognition technology，through the voiceinteraction，the formor manual work can be done easily．DSP embedded systemdesigned in this article can be widely used in handheld command and control devices，with great value and reference value．...