Mahalanobis Distance Calculation Module for Speech Recognition

LI Peng,ZHI Qiang,DONG Ming,LIANG Weiqian,LIU Runsheng
DOI: https://doi.org/10.3321/j.issn:1000-0054.2008.07.035
2008-01-01
Abstract:This paper discusses a low cost,low power-consumption hardware implementation of an embedded speech recognition system.The mapped algorithm,called the Mahalanobis distance,handles about 50% of its computational load in the overall speech recognition algorithm using a continuous hidden Markov model(CHMM).An analyses and tests show that the system can achieve real-time performance at relatively low clock frequencies,which significantly reduces the power consumption.The die area is just 1.2 mm2 based on the HJTC(He Jian Technology Corporation) 0.18 μm technology,including 64 kb SRAM(static random access memory).As a result,this module significantly improves the performance of the whole speech recognition system to a low cost,low power system.
What problem does this paper attempt to address?