Make my PC text to me!
April 8, 2008 2:36 PM   Subscribe

Wave to Text? How can I convert .wav files to text free or cheap? Vonage charges $.25 (each) to convert voicemails to text then SMS them to me. I'd like to do it for free.

I can handle automatically downloading attachments and automating sending SMS messages (I think), I just need software to convert the messages to text. Does anyone know of any good free software to convert them?
posted by TheDukeofLancaster to Computers & Internet (13 answers total)
 
If you're leaving a voicemail for yourself, jott might work.
posted by ejaned8 at 3:28 PM on April 8, 2008


Uuencode and uudecode are the old-style unix way of doing it. Nowadays, yenc encoding is popular because it produces smaller encoded files. Versions of these programs exist for the PC.

You may want to consider converting your wav files to MP3 files if the amount of SMS text is important.
posted by DarkForest at 3:30 PM on April 8, 2008


I don't know of any free, finished software that will do this. The problem is hard, and the people who've solved it pretty well want to be rewarded for it.

However, if you're a programmer, you could use Sphinx-4 (google it) as a voice recognition engine. It needs a proper dataset that phonetically describes each individual word that you want to recognize, and you'll need a badass processor to pull off such a huge dictionary of words, but you can do it with sufficient work.

Then, having done that sufficient work, you'll want to be paid $0.25 apiece for this service.
posted by Netzapper at 3:32 PM on April 8, 2008


DarkForest: I assume the OP is looking for a way to get the spoken words in the .wav written out, i.e. speech-to-text, as opposed to converting the binary .wav file to an ASCII format so that it can be reconstructed bit-for-bit as a .wav on the other end.

I haven't looked in ages, but I'm not aware of any software that actually does this well.
posted by fogster at 3:38 PM on April 8, 2008


Best answer: Dragon NaturallySpeaking from Nuance will process WAVE audio into text for you. It's not free, but it's a one-off cost. About $250 for Preferred (though check the variant you buy supports audio to text.)
posted by alasdair at 3:43 PM on April 8, 2008


Best answer: Yes, I see now. Ignore my comment. I doubt you'll find very good quality speech to text without spending a lot and having very good audio, and spending time training the software.
posted by DarkForest at 3:47 PM on April 8, 2008


I would say 25 cents is very cheap for someone to listen to the message and type it out for you, which is almost certainly what Vonage is doing and what you would have to do to get good results.
posted by kindall at 4:44 PM on April 8, 2008


Best answer: It'd be pretty cool if you built it on Amazon's Mechanical Turk infrastructure. Listening to and transcribing a 30-second phone message is something somebody could easily do in a couple minutes, if they were a native speaker but weren't trained.
posted by Netzapper at 4:56 PM on April 8, 2008


Response by poster: @Netzapper: You're right, how silly of me to think that someone may develop software for free. Please excuse me while I contact Sun, Mozilla, IBM, Asterisk, Canonical Ltd, etc to inform them of their folly.

@Everyone else. Thanks! I like the idea of the mechanical turk, however it does feel like a bit of invasion of privacy to have someone actually reading them.

@kindall During their free trial I had about half the messages with indecipherable words in them. Almost 25% of the text messages were useless either because they simply said I had a voice mail they couldn't transcribe, or because key words were replaced with "(??)" where their voice recognition software couldn't transcribe the messages.

Since I pay $.10 per text received to my phone anyway, this brings the cost for me up to $.35. Over all, it's not a huge sum of money, however it's enough for me to not want to pay for the privilege (yet).
posted by TheDukeofLancaster at 7:09 AM on April 9, 2008


Aha. My error was in expecting that the Vonage transcription was any good. I should have known better.
posted by kindall at 8:35 AM on April 9, 2008


I second ejaned8's recommendation of jott

It's free (for now) and fairly accurate if you can live with its limitations (primarily message length and US telephone number for input)
posted by dudeman at 11:40 AM on April 9, 2008


you are basically looking for a voice recognition program.

The best one on the market is dragon. You can buy the standard version on ebay for 50 bucks or less. I did. To save money you can get an older version too (though I woundnt get anything below version 8, which was the first very highly acclaimed version. I think they're at version 10 now).

In general Dragon works incredibly well. However, it is very picky about the quality of the input. If your wav files are at all noisy, or if they havent been recorded on very particular voice recorders, chances are it will simply not work for you. Ordinary voice recorders wont do. On the dragon website they list 'approved and tested' voice recorders, pretty much only those on the list will work. In other words you really cannot give Dragon any old random wav file and expect it to work. (I mean if it does, it would be pure coincidence and luck). I've tried Dragon with a variety of wav sources and for me only those recorded on approved voice recorders did the trick. If you get an approved one however, it really does work amazingly well.
posted by jak68 at 4:43 PM on April 9, 2008


Spinvox - got a free trial about 8 months ago and it's been working great. I'm in the US but from a first glance at the website it's a little unclear how/how much to activate in the US.
posted by jbradley at 7:37 PM on April 12, 2008


« Older Usenet dwellers, lend me your expertise.   |   Career change dilemma Newer »
This thread is closed to new comments.