System-on-Chip Architecture for Speech Recognition

This paper proposed a system-on-chip (SOC) architecture for speech recognition which is speaker dependent. The feature extraction bases on LPC (linear predictive coefficient)-cepstrum coefficients, and template matching employs Hidden Markov Models (HMM). It does not aim to offer a sophisticated solution but rather a high speed solution. This SOC architecture includes an ASIC of LPC-cepstrum and a Dual-ALU processor. The proposed ASIC of LPC-cepstrum can reduce the calculation load of processor in the speech recognition system. To reduce the area of this ASIC, the resource sharing method is adopted into our design. In addition, this paper also proposed the Dual-ALU processor which provides parallel calculation capability. Hence, it can run more complicated algorithm of speech recognition. For the consideration of chip size, the area of the second ALU is only half of the first ALU. From the experiments, the speech recognition system can provide a high speed solution.