from Hacker News

Ask HN: What is the best off the shelf speech recognition solution

by xb on 5/5/15, 6:18 PM with 1 comments

I am interested in transcribing audio files to text. I don't need 100% accuracy, but the more accurate the better. I have googled around a bit but haven't found anything that worked the way I needed it to. I tried a few free web APIs that worked for short bursts of text but did not hold up for longer audio files. I also have looked into Sphinx, but I haven't gotten past the setup phase.

Does anyone have any experience transcribing long audio files to text, and if so can you share your experiences with the different software solutions you tried?

  • by alt_ on 5/5/15, 6:46 PM

    Artificial artificial intelligence is still your best bet.

    My company does a ton of transcription and we tried a bunch of automated solutions, but they all failed for normal speech. They can catch certain phrases and attempt to interpolate, but it always turns into a jumbled mess when the talker stops enunciating perfectly (<10% accuracy).

    Human-powered transcribing services cost a few dollars per minute and give quality assurances.