Do all voice-to-text algorithms operate in real-time?

Question

Do all voice-to-text algorithms of current technology operate in real-time? I don't mean with a person sitting at a computer with a microphone, but rather inputing a pre-recorded audio file.

i.e. If you have a 30 minute voice recording, will it always take 30 minutes to be transcribed?

Are there different approaches out there?

Answer 1

There is no reason why speech recognition must take as long as the length of audio to be performed. However, due to the computation required I don't think that you won't get a hugely faster than real time algorithm. See this section of the Wikipedia article for more detailed information (it doesn't seem to give any times, though it does give a decent overview of performance).

Answer 2

There is nothing stopping the algorithm running faster than realtime. The Naturally Speaking 10 Professional software program provide a "transcribe from file" option for converting dictation taken while away from a computer, this operates as fast as the computer on which it is running can achieve.

I believe batch processing implementations exist in the area of signals intelligence but such programs would, naturally, be unavailable to the general populace.