Using knowledge of children's speech processes and an expert-based rule system with automatic speech recognition for children's speech
Shelley I. Miller, Utah State University, United States
Utah State University . Awarded
Speech recognition technology is successfully being used in many current applications. One exciting future application is its inclusion in computer-assisted instruction, particularly for children under the age of 10. However, all recognition systems are based upon adult speech. The unique problems that children's speech poses have been largely ignored.
One way in which children's speech differs from adult speech is articulation. This project used a knowledge-based method that derives its knowledge of children's articulation differences from the speech processes of phonological process analysis. The phonetic definitions of words that represent possible developmental articulation errors are mapped from the speech processes by an expert-based rule system. The knowledge-based system uses these rules to generate dictionaries, which contain the developmental articulation errors. The generated dictionaries can then be used with a speech recognition system.
How well the rule system mapped the articulation differences from children's speech processes was evaluated. Adult and child speech samples were tested for recognition accuracy using standard adult-based articulation dictionaries. There was a significant difference in the recognition accuracy between the children and adults. To examine how the articulation differences affected recognition accuracy, two types of dictionaries were used with the speech recognition system. The exact transcriptions, determined by a speech pathologist, of each child's speech samples were used for recognition testing. The generated dictionaries containing the articulation differences, representing each child's speech processes, were used. Both approaches showed no significant increase in recognition accuracy.
By using the articulation error pronunciations returned by the speech recognition system, the speech processes that the speech recognition system prefers were found. Recommendations, arrived at by comparing the speech processes preferred by the recognition system and the recognizability of the test words, are made that may be useful in developing or modifying a speech recognition system for children's speech.
Miller, S.I. Using knowledge of children's speech processes and an expert-based rule system with automatic speech recognition for children's speech. Master's thesis, Utah State University. Retrieved February 20, 2019 from https://www.learntechlib.org/p/119030/.
Citation reproduced with permission of ProQuest LLC.
For copies of dissertations and theses: (800) 521-0600/(734) 761-4700 or https://dissexpress.umi.com