The word I hear is that the results are mixed. In discrete speech you are
mapping an utterance directly to a word, phrase or macro. Therefore the
quality of that utterance is linked only to the user's ability to recreate
the utterance to accomplish good recognition. Discrete speech product
don't care much about specific characteristics of an individual's speech.
In a continuous speech environment the voice models are more precise and
the inter-relationship of the utterances and how they combine can be
particular. Users with a strong accent and children or adults with unusual
speech patterns sometimes will have difficulty with yielding excellent
continuous speech accuracy. In some cases users with a quiet voice or
compromised breath support can offer a challenge. The problem is that
there are few absolutes that we have been able to determine. We are
surprised often enough about the results that I encourage you to just give
it a try when working with an individual that could benefit from this
technology. We have seen more than one speech pathologist eat their words
after stating that speech recognition wouldn't work for someone with
impaired speech. We will often do an evaluation for users we feel may have
a problem with speech recognition. With NaturallySpeaking you can train a
user voice file in 30 to 40 minutes and give it a shot. When it works
properly (and it usually does) the results are clearly worth the effort.
If you end up doing much work with children and continuous speech I
encourage you to publish your results here.
Regards,
Marty Tibor
Synapse
3095 Kerner Blvd., Suite S
San Rafael, CA 94901
Phone 415-455-9700
Fax 415-455-9801
toll-free 888-285-9988
martytib@synapseadaptive.com
http://www.synapseadaptive.com
Providers of adaptive and assistive technology solutions.
At , NaIcke at Internet wrote:
>Marty,
>
>It was not just the training component of Nat Speaking with students that we
>were worried about (although that was one issue), I was told by someone
>within Dragon that it would not accommodate children's voices. Do you
know if
>this is true or not? I would be very interested to find out as we have many
>students who are currently using discrete speech and would like to try
>continuous speech. Please let me know if you have any information on this.
>
>Thanks Nancy
>
>
This archive was generated by hypermail 2b29 : Tue Jan 04 2000 - 12:43:29 EST