Speech to text without a secretary

December 29, 2005 10:07 PM

Is there a free/cheap speech-to-text converter (ie. dictation, verbal notes) that works reasonably well?

Best if it accepts wav files. Bonus points for Linux command line, but Windows is acceptable. All I really want is raw text from speech-only audio, and it doesn't have to be perfect. Oh, and I don't want to invest massive time cobbling together my own tools from the half-baked libraries and things that are out there.

posted by Kickstart70 to Technology (8 answers total)

Response by poster: And heck, I suppose even if it's not free/cheap I want to hear about it, though f/c is better obviously.
posted by Kickstart70 at 10:12 PM on December 29, 2005

I've looked for free/cheap, and they are so terrible they have more entertainment value than MadLibs. I'm not posting this to be noise, but rather to be helpful in that even the "best" voice-to-text programs are going to be pretty bad. On one program, I tried speaking "I need milk from the store" and it came out something absurd like "Beat doorframe on tap." Just to save you some aggravation in playing with the free stuff - though if you have time to kill, it is fun.
posted by Iamtherealme at 10:24 PM on December 29, 2005

Not cheap per se, but: if you already have MS Office installed, it comes with a voice-to-text tool. I believe this started with Office XP, and was definitely part of Office 2003.

It may have even come with the stand-alone version of Outlook.

As for efficacy, I think it qualified as "pretty good". Not as good as the two dedicated/high-end packages around at the time (IBM had one), but better than the cheap/free stuff.
posted by Dunwitty at 10:46 PM on December 29, 2005

As far as I know, even the very best ones out there require several things that plain recordings may not provide:

1. A very clear and noise-free microphone
2. Some amount of "training" for the program to recognise the particulars of the voice by reading a long series of specified words/phrases to it
3. A slow and deliberate pace that is much slower than natural speech, so that the program can have time to process each word

Now obviously these are not insurmountable to overcome, and heck there might be speech recognition programs out there that are designed to work in this mode, but I think most of them are meant for live dictation where the speaker is sitting at the computer and can go at the proper pace and where the program has been trained.
posted by Rhomboid at 10:48 PM on December 29, 2005

my father has had good success with Dragon Naturally Speaking, though it is a fairly complex product.
posted by chefscotticus at 5:49 AM on December 30, 2005

This is neither free nor cheap and requires a phone call be made, but it's something to throw out there...call copytalk. Disclaimer, I used to type for them and some of the stuff we had to type was really tough to understand, but we muddled through it as best we could. So again, you have to be a clear speaker for it to work. www.copytalk.com. There are different levels of talking (different lengths of messages, they don't let you go on an and on forever) at different price levels.

I've never used it, so I can't speak to the turn around time but the customers seemed pretty happy.

let us know what you decide to use!
posted by bilabial at 8:20 AM on December 31, 2005

I'd send an e-mail to your local university's assistive technology office (or whatever they call it there) or one of their special education faculty members. In my experience academics like to share info about neat things they've "discovered" like that sort of software.
posted by BrandonAbell at 12:01 PM on December 31, 2005

Samsung had a cellphone with the ability to do this for text messages etc... i'm not sure how well it worked...

200 investment to try though.
posted by stratastar at 8:23 AM on January 3, 2006