On Thu, 7 Jun 2012, T L wrote: > On Wed, Jun 6, 2012 at 9:04 AM, Raymond Norton <admin at lctn.org> wrote: > >> I need to transcribe from voice files to text, with a variety of >> voices in the various recordings. Been googling around and not finding >> a solution I am happy with or suits my needs. Any recommendations? > > > Sometimes the right tool for the job is a human. > ;-) > > Toss this into google: transcription from recording outsource > > Here's the thing: speaker independent speech recognition is a hard > problem. It gets harder with multiple voices. And, unless your > speakers were talking directly and clearly into the mic, you're likely > to have non-optimal sound quality. This, too, makes it harder for a > computer. > > You'll spend a LOT of time doing clean-up after the fact, if you don't > have a human do it in the first place. I use Google Voice for voicemail and forwarding phone calls. It's a pretty brilliant scheme and I recommend it. One thing it does is transcribe messages to text and forward them to me via email. It does a good enough job that I can almost always tell what the message is about, at least, but it usually has some errors, often laughable ones. It can also record calls and transcribe the conversations. There is no doubt that you would have a lot of work to do fixing up mistakes in Google Voice voicemail transcriptions, but I would find them very helpful if I were trying to transcribe recordings. If I were a very fast typist (typing as fast as someone can speak), then I'd prefer to do it by hand. I first used voice-recognition software back in about 1999. I think it was from IBM and it was supposed to learn my voice and improve over time. It wasn't too bad, even at first, but I didn't really have much use for it. I've been thinking that with improvements in software and massive improvement in computing power, today's voice recognition software should be great. Maybe that only applies to cases where the speaker is trying to be understood by the software. Casual speech is always going to be difficult to understand. Mike