About DTW-algorithm capabilities for speech recognition.
Nitsenko A.V., Shelepov V.Ju.
In the article are considered variants of application for various problems of the speech recognition method based on the DTW algorithm using phonemic segmentation and word templates automatically synthesized from the diphone templates. The method advantage is that it allows to apply the DTW algorithm for large vocabulary speech recognition (more than ten thousand words), as well as for continuous speech recognition. The use of the speaker’s diphone samples database together with the automatic transcription makes it possible to work with arbitrary dictionaries specified as the text by synthesizing the word templates from diphone templates without the need to perform the preliminary training of the whole word templates.
