[asterisk-dev] Speech Recognition

[Kevin P. Fleming]

----- John Lange <j.lange at epic.ca> wrote:
> A "sort-of" on topic question here. Are there different speech
> recognition engines depending on the application?
> 
> It occurs to me that speech recognition might be much more accurate
> if
> it wasn't required to do its analysis in real time. So for example,
> transcribing voicemail to text could be done off-line after the
> caller
> hangs up.

The simple answer is yes.

The more complex answer is that in fact those are very different applications; an engine designed to be able to quickly and accurately recognize phrases is very different from one designed to accurately (but not necessarily quickly) recognize free-form speech. Even the existing engines, like LumenVox, will start to produce poor results if you give them a grammar that is too large or has too many overlapping/similar sounding phrases.

-- 
Kevin P. Fleming
Senior Software Engineer
Digium, Inc.