Improved Accuracy of Automatic Speech Recognition System using Software and Hardware Co design

We Present Automatic Speech Recognition System For Human Machine Interaction By Using Hardware and Software Co design, The Entire Speech Signal Is Segmented Into a sequence Shorter Signal Is done By Future Extraction, This System Improves the Timing Performance by using Hardware Accelerator GMM (Gaussian mixture model), This hardware accelerator is performed by emission probability Calculation, In order to avoid large memory requirement and this accelerator adopts a double buffering scheme for accessing the acoustic parameters, The weighted finite state transducer (WFST) in Viterbi search model gives Word Sequence from Emission probability calculation. The word accuracy rate is preserved at 93.33%, as a part of recognizer is very suitable for Embedded Real time Application By Using New Adaptive Beam Pruning algorithm. Which is about four times faster than pure Software based system.