A robust speaker-independent speech recognizer on ADSP2181 fixed-point DSP

We implement a robust speaker-independent speech recognition algorithm on an ATSP2181 fixed-point digital signal processor (DSP). The recognizer is robust against environmental noise, such as air-conditioner noise, computer-fan noise, and keyboard-typing noise, etc. We address several implementation issues to ensure that errors due to truncation, which are inevitable in a fixed-point implementation, are within an acceptable level. Preliminary experiments we conduct show that our DSP-based speech recognizer attains at least 95% recognition accuracy in a noisy environment