DSP Implementation of Speech Recognition Based on Endpoint Detection of Speech

The author proposes one algorithm that detects the beginning and ending boundaries of speech based on dynamic estimating environmental noises.This algorithm can work well with more accurate and robust than time-domain-energy-based endpoint detection.A speaker-independent speech recognition system that contains endpoint detection of speech is developed based on TMS320VC5410 Evaluation Module(EVM).The experiment based on isolated-word speech-recognition shows that:this algorithm contributes to the increase of recognition rate.