The recognition system for real world must have robustness to noise and rejection capability for false inputs. In this thesis, the isolated word recognition system is designed with ZCPA(Zero-Crossing with Peak Amplitude) model based on human auditory system and RBF(Radial Basis Function) networks which shows the high recognition rate in the small isolated word recognition system and rejection performance.
The filter banks of ZCPA model are composed of FIR filters with powers-of-two coefficients. This filter needs less area. Time and energy normalization are performed for the ZCPA feature vectors. The 1024 input nodes, 50 hidden nodes, and 50 output nodes is used for RBF networks. Efficient and powerful memory operations are adopted. RBF networks are realized using pipelined Finite State Machine.
ASIC(Application Specific Integrated Circuit) chip for speech recognition is designed for real world application and verified using the delay information from the layout. Designed speech recognition chip operates on real time, and shows good performance.
