|
|
 | | From: | deltagreen | | Subject: | Re: Mac OSX | | Date: | 14 Dec 2004 08:09:25 -0800 |
|
|
 | Creating special pronunciations where a word is represented by a
grammar of succession of single phonemes is risky, but may be worth a
try. You may then end-up with a successful grammar match and the times
associated with the recognition.
As a general rule though, in today's speech recognition, you should
stay away from phoneme recognition level. Some people needs it though,
like for drawing lips to a TTS character - but that's at the TTS level
where they know which phoneme is being spoken at all time since they
drive the process. On the speech recognition side, rare are the cases
where you need to know exactly which phoneme is being spoken at every
given time.
I am not aware of what you are trying to achieve, but there's a red
flag if you need the phoneme time and score for your recognition.
Phonemes are nothing in speech if it's not to get to recognized
pronunciations.
|
|
| | From: | Richard Owlett | | Subject: | Re: Mac OSX | | Date: | Tue, 14 Dec 2004 13:56:52 -0600 |
|
|
| deltagreen wrote:
> [ possibly significant snip ]
>
> As a general rule though, in today's speech recognition, you should
> stay away from phoneme recognition level.
Can you supply non-technical supportive link(s)?
I'm interested in "command and control" and/or "limited vocabulary"
applications.
I had a friend who was involved in speech recognition research back in
early 70's. He was thrilled to have 1/2 of a KL-10.
I casually followed some SR in late 70's / early 80's ( discrete rather
than continuous )
My 'gut feel' is "state of the art" has REGRESSED :(
OK, so ' AND " are significant ;)
But I think I somehow represent one potential end user class.
BUT, I also recognize that I may not know enough to pose useful questions.
Please refer me to WEB resources as I *DO NOT* have access to any
academic library.
|
|
|
