A three stage hybrid model to perform feature level speech signal recognition

In this paper, a three stage improved speech signal recognition model is presented. The presented approach improved the recognition process by reducing the process time and to provide robust speech recognition. In first layer of presented model, the feature extraction from speech is done using Statistical Analysis based DWT approach. The extracted feature based recognition reduced the signal size for analysis. At second stage, the signal filtration is performed using spectral subtraction method. These two stages, converted the signal to normalized form on which two levels HMM method is applied for recognition. The obtained result shows the 90% accuracy rate in effective time.