LOW-RESOURCE REAL-TIME SPEECH RECOGNITION SYSTEM USING AN OVERSAMPLED FILTERBANK

摘要

A low-resource real-time speech recognition systemdigitises voice signals, applies a pre-emphasis window tothe digitised voice signals and stores the result,calculates FFT coefficients, factorizes the band energiesduring the process of MFCC calculation, and performs allother operations needed for speech recognition. Thespeech recognition algorithms are implemented on an ultralow-power, small-size, and low-cost platform in realtime.