How do I get the text for live closed captioning for church sermons?

September 9, 2013 12:17 PM

I have a church and they have a signer for the deaf. They have asked for live closed captioning, but they do not want to have someone type it in in real-time. Is there speech recognition software good enough for live speaking, or even something that handles closed caption-style graphics? Once I get the text, I can handle it from there with some kind of video overlay. Thanks!
posted by ostranenie to Media & Arts (5 answers total)

I'm pretty sure that voice recognition technology isn't accurate enough these days that you'd want it going live in church. If you want to see some of the best in action, Google has some auto-detect technology built into YouTube for closed captioning where it hasn't already been typed in, and it's pretty hilarious how off it is sometimes.
posted by SpacemanStix at 12:32 PM on September 9, 2013

For what it's worth, broadcasters often do this with a "respeaker": someone who listens to the audio and repeats it at once into a TTS system. The idea is that the respeaker is deliberately speaking very clearly in a quiet environment with a good mic, with software that's been trained against their voice, which makes it much more reliable than just throwing the original audio at a TTS system. AFAIK it's still less reliable than a skilled steno operator though. More detail here.
posted by pont at 12:59 PM on September 9, 2013

Is there a specific reason they don't want to do live-time CART services? I know a CART typist who freelances for a TV news channel -- I believe there's a software that transcribes it, and then the typist corrects it, but the baseline accuracy is so bad that she ends up typing most of it anyway. I myself am not a churchgoer, but every deaf/HoH person I know who is religious goes to churches with CART typists and my understanding is that it's quite common.

Anyway, if this is meant for real time rather than a later taped broadcast, also keep in mind that the delay for transcription + correction may be a few seconds. For the caption-style formatting, look at *.sub or *.srt formats.
posted by angst at 1:29 PM on September 9, 2013

Re: Realtime CART services, they may not want to pay someone to transcribe stuff real-time (hence asking after a robot to do it automatically for free), but if they do, that looks like The Answer.

Also, difficulty level: Latin, so I doubt a realtime voice recognition solution will work well anyhow. Thanks for the answers.
posted by ostranenie at 2:22 PM on September 9, 2013

Oops, obviously I meant voice recognition when I said TTS. And you're right, for Latin I think voice recognition would give very poor results.
posted by pont at 11:40 AM on September 10, 2013