Low power design techniques for speech recognition

Recent research focuses on low power design techniques. This has been mainly motivated by the demand of hand-held electronic devices which must consume less power. This thesis presents a hybrid approach for designing a low power Multi-Layer Perceptron (MLP) based Neural Network (NN) for speech recognition. They are bipartite tabular method and banking organization method. The MLP based NN is trained in Matlab using TIDIGITS corpus. This approach is simulated in Xilinx xc3s1200. The system is evaluated using optimized model weights which are exported from Matlab. Performance parameter like area and power is computed.