I need to find a transcription service that can transcribe ~10k very short recordings.

February 22, 2011 10:41 AM

I need to find a transcription service that can transcribe ~10k very short recordings. All different speakers calling from various places. I know of Casting Words and am trying to find out if they can do this. Each recording may be from one to a few words. Perhaps a date. Can you recommend any other resources I can pursue? Thanks.
posted by reddot to Computers & Internet (10 answers total)

Casting Words does alot of their work (if not all) through mturk, the amazon service. You could look at mturk to see similar requests by various people.
posted by quodlibet at 10:48 AM on February 22, 2011

2nding Mechanial Turk. Here's a good guide.
posted by k8t at 11:01 AM on February 22, 2011

I've posted requests for bid for transcription jobs on Elance, and the results are consistently excellent. There are tons of people on there who specialize in it.
posted by jbickers at 11:17 AM on February 22, 2011

Response by poster: I've thought of using mturk directly, but I don't want to have to manually verify each of the 10,000 transcriptions. Unfortunately, I don't have any programming resources to assist me.

I've used mturk directly for really small jobs but am daunted by the size of this one.

I'll ckeck out elance.

Any further suggestions would be appreciated.
posted by reddot at 11:51 AM on February 22, 2011

Craigslist, two jobs: 1) the person to transcribe the recordings, 2) the person to check the transcriptions.
posted by zippy at 12:05 PM on February 22, 2011

Best answer: Along the lines of zippy's suggestion, why not run a couple mturk jobs? Depending on how much you want to spend and how much you trust the mass of "turks," you could have two runs of translations and two of verifications? You could still come out at a lower cost than other services (k8t's link adds Purple Shark and Kedrowski Transcription to the list of transcription services).

Or, make playlists for these, and do a few minutes between other tasks in your day. How long is the whole list, time-wise? Set up an audio-playback program to use universal hotkeys that allows you to pause when you need, or to not play continuously.
posted by filthy light thief at 12:19 PM on February 22, 2011

Elance, vworker, do random spot checks. First do 500 recordings, spot check, if the worker is good, keep giving him recordings in 500-1000 sets and spot check 5-10% of them as you go along. Much easier to get consistent quality than mturk.
posted by rainy at 12:19 PM on February 22, 2011

I used to work for these folks; they are expensive, but have very low error rates and quick turnaround.
posted by Sidhedevil at 6:17 PM on February 22, 2011

Response by poster: I'm following up the leads given here. Thank you everyone.
posted by reddot at 2:50 PM on February 23, 2011

Just as an aside, there are statistical sample size methods you can use to develop a level of confidence for things like this so that you don't have to check every single one, and the results can be defensible, if it matters to you. You can find a good overview here. If you'd like help running some numbers given your specifics, I'd be more than happy to help - I'm an auditor and love stuff like this. :)
posted by TheNewWazoo at 8:30 PM on February 24, 2011