[LMB] OT: Voice transcription software
mark at allums.com
Thu Sep 16 04:42:19 BST 2010
I think you're right, Ed, but in the past, no amount of training was
enough, and also, enunciation is my weakest spoken word trait. So, I am
at a disadvantage, I guess. Maybe with computing power going to waste
these days, they've managed to make it work. I'm glad; this technology
is badly needed. Good luck to everyone who tries it.
On 9/15/2010 9:52 PM, Ed Burkhead wrote:
> In general computer speech recognition doesn't have so much problem with
> accents - that's what the training takes care of.
> What it needs is ENUNCIATION.
> For example, when in Germany, watching German TV, I was astounded at the
> mouth movements of all the actors. Since there's such a difference in
> the language from Eastern Austria up into the Netherlands, many people
> can't understand the actors unless they overdo the ENUNCIATION. It is
> funny watching them, especially the kids, making huge mouth movements to
> speak SO CLEARLY.
> As I coach speech contestants, I continually emphasize enunciation. If
> the judge and audience can't understand your words, because they haven't
> heard all the separate phonemes, then you might as well have stayed
> home. You might try this same thing on the computer - exaggerate your
> speaking to the point of saying E-NuN-Cee-ATe. As you master it, you can
> probably get faster and faster and tone it down, some.
> Perhaps it's kind of like giving commands to dogs. Dogs don't, I've been
> told, tell vowels apart. When you tell them to sit. It needs to be SiT
> with a hard, final T. When it's come, it needs to be a hard Kom.
> Letting the computer program hear all the phonemes lets it correctly
> identify the words.
More information about the Lois-Bujold