Looking to convert a recorded class's MP3 into text via speech recognition

April 20, 2010 12:58 PM

I would like to run MP3 files through speech recognition to turn them into text (recorded class lectures). Can anyone recommend software to do this (directly from the mp3's - I can only find speech recognition for microphone input)
posted by lrodman to Computers & Internet (11 answers total)

can you find a program that would let you choose your input and run the speaker out to the line-in of your computer? I'm not sure if that would cause the universe to explode or not.
posted by CharlesV42 at 1:22 PM on April 20, 2010

Or just play the MP3s - I'm assuming speech recognition has to work in real-time?
posted by djgh at 1:33 PM on April 20, 2010

Here's a video showing someone doing that using Dragon NaturallySpeaking software.
posted by davcoo at 1:36 PM on April 20, 2010

You may not get acceptably good results using transcription software. You might try a transcription service (there are many available online), which, depending on the total length of audio you need to transcribe, could be very cost-effective.
posted by brain at 1:48 PM on April 20, 2010

I don't think speech recognition would work very well on this type of project, because it would be trying to convert the musical instruments into words as well. plus it most likely wouldn't understand someone singing words very loudly or not pronouncing them very well...especially when covered by distortion and things of that nature. If it's relatively calm music with a distinct voice then you may have a shot.

Dragon Naturally Speaking is typically the go-to software for speech recognition, but it's not cheap...(i.e. $200)...and if there's a good chance that it won't work, then you may not want to go that route.
posted by AltReality at 2:27 PM on April 20, 2010

Have you tried speech recognition software with your own voice? In my experience, there is a lot of "teaching" the software and manual editing involved to get to the point where it recognizes a person's voice. It's not a set it and forget it type of deal.

If you want to try it, just plug your audio player into the microphone port you normally use and test it out.
posted by reeddavid at 2:30 PM on April 20, 2010

Adobe Soundbooth is supposed to do this, even sorting different voices if I recall correctly, but I haven't tried it myself. There ought to be a 30 day trial of the software though.
posted by monocultured at 3:15 PM on April 20, 2010

I am writing this post using Dragon NaturallySpeaking. It takes a lot of training and it is not perfect even with a good microphone and a quiet environment. You will need a very expensive recorder to make this happen and at least the preferred version of the program. even then given the acoustics of the room and the background noise, the lack of training required for the sound profile to produce good results, and the distance from the speaker is highly unlikely to produce very good results.

additionally, because you won't be able to train the program using the speaker's voice, you will have to use a product called univoice, which adds to the expense even further.

The go to website for this kind of information is KnowBrainer speech recognition forums. They also have the products needed.

just for your information, this post was made during my lunch break in a an office environment and included at least seven errors despite the best microphone and extensive training of my voice profile.
posted by chinabound at 10:02 PM on April 20, 2010 [1 favorite]

I hear that Amazon Turk does a lot of podcast transcription...though I've never quite figured out how to utilize Amazon Turk. But it's very inexpensive and from what I've read it's far more reliable than speech-to-text software.
posted by arniec at 7:16 AM on April 21, 2010

Response by poster: I tried dragon but it requires training, it didn't work even remotely.
posted by lrodman at 8:32 AM on May 2, 2010

Response by poster: are you sure adobe soundbooth does speech-to-text?
posted by lrodman at 8:34 AM on May 2, 2010