Implementation of speaker-depended speech recognition algorithm in VC++

In view of problems of huge words database and complex training process existed in current speaker-independent speech recognition system, a speaker-depended speech recognition algorithm was implemented by using VC++. The algorithm uses method of pre-emphasis first then endpoint detection to eliminate influence of low-frequency noise; selects two indexes of short-time energy and short-time zero-crossing rate as criterion of endpoint detection for speech signal; adds judgment of tolerable mute time to ensure integrity of speech data; uses improved dynamic time warping algorithm to make pattern matching which reduces computation under premise of not influencing calculating results. The test result shows that the algorithm can recognize short words and short sentences real-timely and accurately with faster recognition speed.