I have many hours of recordings of group discussion. For the most part, each person talks separately from the others in time, and there is minimal background noise. I need to be able to find instances of one voice (or each voice), and either mark these, or create separate clips from them.

Ideally, I would set one voice as the key voice, and have some process identify when that voice recurs. I have a lot of audio to use for comparison.

I have access to Pro Tools 12, Reaper, and many audio processing software packages for music and noise reduction.

Any ideas?

  • 1
    There is no shortcut to this - you will need to transcribe the recordings against timecode and then identify the regions of speech that are useful to your project. – Mark Nov 27 '16 at 13:19

Marcus Weaver-Hightower, PhD states in this video that "there's no magic transcriber out there" to remove human effort from the transcription process. That being said, his suggestion of using voice recognition software to speed up the transcription process may be something of value to you.

You could consider using software like the Max Planck Institute's transcription software ELAN if you need your transcriptions to be accurate or complex, but as with most freeware, the learning curve and user-friendliness can be lacking.

With regard to software that can serve up the magic bullet you are seeking, I have not encountered such a program, but I will concede that such a program would be theoretically possible. Since I imagine you probably don't have the time or skills to code such a complex program as that would be, I think your best bet is to transcribe manually, possibly by using Dr Weaver-Hightower's suggestion.

  • Time and ability are the real issues here. It is an ongoing problem, so it might be worth investing in, but the uncertainty of results from my own efforts makes significant investment seem foolish. – Shawn Dec 15 '16 at 16:23

The only thing to do here is edit it and put individual voices on their own tracks. In order to have the processes occur on the correct voices each track will have its own inserts and sends, so you simply have each track with its own processes. You could use automation but I think that's a retarded way if working.

  • I'm asking about ways to speed up editing and/or finding points for edits. Any thoughts on how to go about that editing without having to listen to the multiple hours of discussion? – Shawn Nov 25 '16 at 15:44
  • 1
    No. The only way to speed it up would be if you had a script with timecode cues. A DAW, even Pro Tools which has arguably the best editing tools simply cannot know 'where' a specific something is without it being told to go there, it just reads back the digital audio. There is no tool that can auto identify different voices simply from the audio. – Melloj Nov 25 '16 at 16:56
  • 1
    There is no way of isolating individual voices to place them on isolated tracks unless they were recorded in this way in the first place. A group discussion, unless scripted, will have overtalking so standard dialogue editing techniques will not be possible in this case. – Mark Nov 27 '16 at 13:17
  • He indicated in his question that the voices for the most part are separate, so some traditional editing would be possible. However he was asking if a DAW could identify different voices on the fly using (I would suggest amplitude and frequency) to auto detect but no DAW does this. – Melloj Nov 27 '16 at 13:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.