Dynamic Time Warping (DTW) in small vocabulary word recognition.

摘要

In recent years speech recognition systems have been used for practical applications such as assisting disabled people, automated services through telephone, and other applications. Many applications are based on the use of a limited vocabulary. Presented here is the development of a small vocabulary word recognition system and its characterization compared to other approaches. There are two phases in the study: a training phase and a testing or recognition phase. The Mel Frequency Cepstral Coding (MFCC) technique was used in the training phase and Dynamic Time Wrapping (DTW) was used in the testing phase. The nearest neighbor selection approach was employed with the DTW technique to find the best match. A comparison is made between four speech recognition approaches: DTW based, hidden Markov model based, and two commercial recognition software packages, Windows XP Say Now and Dragon.