April 30, 2001

11:26 AM

Is Speech Recognition Software: What is it good fowah?[caution: link to the vulgar FC]. The apparent demise, or at least fraud, of Lernout & Hauspie inspires me to ask whether speech recognition software can be used to create more than garbage writing, fast. As an attorney, I spent a good chunk of the 1990's trying to get permission from people born in the 1940's to draft my own documents with a keyboard rather than a Dictaphone. Fortunately, I don't think SR programs will ever catch on for more than commanding a computer do something. But maybe I'm completely wrong?
posted by ParisParamus (11 comments total)

And those Dragon Speak folks got ultra-screwed...
posted by ParisParamus at 12:06 PM on April 30, 2001

Dunno how much stock I'd put on a FC article, but here's a Reuters link from news.com that gives a few more details on the arrest.
posted by swell at 12:11 PM on April 30, 2001

Speech recognition has been a godsend to certain people who are severely handicapped, particularly to quadraplegics. People like that whose brains are normal and whose eyes, ears and voices work can leave their otherwise-useless bodies behind and enter the freedom of cyberspace, where they can travel just as easily as the rest of us do.

IBM's speech recognition packages are implicitly intended for such users, because with them a Windows computer can be completely controlled with the voice. No hand movements whatever are required, which is obviously vital for a person whose hands don't work.
posted by Steven Den Beste at 12:48 PM on April 30, 2001

SDB, certainly a noble, worthwhile use. I'm most curious about people's experience with SR as an alternative to a keyboard for writing. I used to feel a keyboard was inferious to a pen and paper for writing; now I'm not sure. Now I feel (as I always have) that I write differently than I speak, and that "speaking" an article or report results either in garbage, or required more so much editing that no speed advantage is gained.
posted by ParisParamus at 1:42 PM on April 30, 2001

I've always wondered about the reliability of products designed for speech to text functionality...mostly because the products are rather expensive and usually have no trial version. Furthermore, any hands-on experience that I've had in the past has been with older, less-developed, gimmicky-type software. Anyone know how products like ViaVoice stand up in, say, "speaking" an email and having it rendered in text? I think that's all I'd ever use it for.
posted by Hankins at 1:43 PM on April 30, 2001

Since you can't see speech, you tend to repeat yourself a lot more often when you speak, and you forget where you're going a lot when you try to dictate long sentences. The products are fairly accurate, but I don't think they're that useful if you can actually type.
posted by kindall at 2:19 PM on April 30, 2001

It goes beyond repeating words. I suspect, even, different parts of the brain are used for speaking and writing. You wouldn't believe how many people still view touching a keyboard as beneath them...
posted by ParisParamus at 2:36 PM on April 30, 2001

I had an executive two companies back who loved to dictate letters to his computer while he strolled back and forth in front of his window. Of course, he had a secretary to clean it up for him afterward, but still.

It's not unheard of.
posted by dhartung at 5:41 PM on April 30, 2001

Typing is for secretaries according to a certain mindset in some middle managements. Some of these guys use secretaries to write their emails for them.

I agree with ParisParamus, speech recognition by itself is of dubious worth to someone who can already type. I'm not against alternative input devices, mice, tablets, gestures and voice etc. All probably have a place but probably not to the extent of completely eliminating the keyboard.

Small mobile wireless applications might be an exception.
posted by lagado at 9:40 PM on April 30, 2001

At work we're investigating various voice-recognition products for use by probation officers in the field. They can dictate chronological entries and other notes into a handheld device and synch it up with the system when they come to the office.

Unfortunately, we haven't found anything sufficient to our needs. Sure, you can train the software to be reasonably accurate under perfect conditions--but in a car or other field location, the stuff is useless.
posted by frykitty at 9:57 PM on April 30, 2001

Stop speculating and perform the experiment.

I just wrote a chapter of a technical book entirely by dictating to Dragon Naturally Speaking (L&H's product). Accuracy is amazing, especially on technical terms that are harder to type than they are to say. I trained the program by doing the narrations at my usual coffee-addled doubletime. Then I fed the program the existing chapters to the program so it'd know my tech terms and common usage. Fantastic accuracy.

Talking and typing are different processes to me. I can construct and buffer the next sentence while typing, but have to think about what I'm saying. It's not entirely clear that narrating a technical book is the best way to generate content.

The downside is when you dictate content the program can't reasonably guess. Going back later for proofreading can be baffling when the text seems to indicate a conversation about the religious aspects of heterodyning monkeys, and you're pretty sure that's not what you were talking about.
posted by glunt at 10:47 AM on May 1, 2001