| As with many computer speech applications, speech-recognition accuracy is one of main problems of the SpeechWeb system and requires theoretical research. Many approaches for improving speech-recognition accuracy have been carried out and some results have been found, but there is still a far way to go. A new approach, which involves embedding semantic constraints in the syntax of the recognition grammars has been developed by Frost. This report describes an investigation of the new approach and other potential solutions. |
