- From: James Salsman <j.salsman@bovik.org>
- Date: Wed, 17 Jan 2001 13:47:00 -0800 (PST)
- To: www-voice@w3.org
Why does VoiceXML still not have speech recognition results including: 1. access to the recorded audio speech recognized; 2. segmentation of the endpoints of words and phones; 3. confidence scores for words and phones; 4. N-best ordering of alternatives in the grammar, even though these issues have been brought up long ago? Does anyone dispute that these functions are essential for educational applications, such as pronunciation assessment and reading skill evaluation? Not to my knowledge; nobody has suggested that the research cited on this list in support of those features was not the very best of its kind, which I believe it still is. Doesn't the W3C and the member companies want to be seen as supporting educational applications of speech technology? On the other hand, I STRONGLY APPLAUD Dave Raggett and the W3C members who have been working on N-Gram language representations. For concrete examples of the reasons why non-BNF grammars are essential for educational applications, please see this ongoing analysis of children's repetitions during oral reading: http://www.bovik.org/reps-char.cgi Cheers, James
Received on Wednesday, 17 January 2001 16:47:07 UTC