Speech link

• Main Authors: Chan, William C., Lim, Lingling N., Lopez, Dennis T., Marfil, Minette K.
• Format: text
• Language: English
• Published: Animo Repository 1993
• Subjects: Automatic speech recognition; Machine translating; Computer software; Information theory; Computer systems
• Online Access: https://animorepository.dlsu.edu.ph/etd_bachelors/6015
• Institution: De La Salle University

Speech recognition plays an indispensable role in man-machine interaction. It is basically a process which allows a person to communicate with a computer by speaking to it, thereby analyzing acoustic patterns of speech signal incorporated into the message sent so that the machine can respond accordingly to the human spoken utterance. Tantamount to this purpose, SPEECH LINK is developed by the proponents in order to recognize human spoken utterance evident in a text-editor format of recognized words. However, this system adapts a limited vocabulary and is speaker-dependent which means that only one person's voice can be used at a time to recognize speech sounds. The recognition process is made possible through the training of word templates generated from the speaker's utterances. These templates will serve as the basis for recognition. The set of all trained words will comprise the vocabulary of the system. Facilitating this process is a software-driven methodology commonly known in speech recognition systems as the Dynamic Time Warping (DTW) technique which is basically capable of matching unknown input utterance to stored reference templates. In this approach, the unknown feature is matched against the stored patterns using a procedure that dynamically alters the time dimension to minimize the accumulated distance scores for each pattern. The word with the smallest distance that is closest to the reference pattern or template serves as the best candidate as the word being recognized. The reliability of the system will be generally dependent on how similarly distinctive the words are uttered by the speaker during the training and recognition modes and the vocabulary size. The software is implemented in the Turbo C language where all the utilities required for training and recognition can be used for further applications.