wired for sound

[doc_strange]

Well, I finally decided to take the plunge and start in on speech to text -- a.k.a. continuous speech -- software. I'm trying right now to see how close to subvocalization I can get it, since ideally one could use it in a quiet room with other people without disturbing anyone. For the moment I'm using Dragon 7.3 on a pretty slow laptop(600MHz).

It's pretty fun but occasionally quite frustrating as it will sometimes smash interesting phrases out of what I'm actually saying. It makes you ridiculously conscious of how you pronounce things, and makes you wonder, "do I really sound like that?"

For example, it has difficulty with very small words used in rapid series. So something like, "what's up with all that?" Might come out as, "it's up with all that" or worse, "what's up with a hat" which is not quite the same thing at all.

But muttering into a microphone and having words -- mostly accurate -- appear before you has a cachet all its own.

This posting for example was typed entirely through the software: only about 10 corrections had to be made. That's not much worse than my typing.

Comments

[tanac.livejournal.com]

Your error rate is probably better than mine, then.
"What's up with all that?" Comes out just fine the first-time though. Then again, I've been using and training the software for several months now. I think what I need to do to get a better error rate is invest in a better microphone.

[docstrange.livejournal.com]

Definitely. What's funny is that the best-rated mic (by Scansoft) is nowhere near the most expensive. The Telex H-841 headset mic with USB adapter (use it, makes a big difference) is something like $46 grand total, shipped from Provantage (who, by the way, rock). That's amazingly cheap when you consider that a normal Plantronics headset runs over $100. Scansoft rated it most accurate by far. My REAL problem right now is getting myself to not chuff into the mic...