Hey everyone! 👋 I'm trying to use Otter.ai to transcribe some internal Zoom training sessions. The audio quality isn't great—lots of background noise, people talking over each other, and sometimes the speaker is far from the mic.
What settings or tricks do you use to get the most accurate transcript from a messy recording? I've tried uploading the .mp4 directly and also the separate audio track. Is one better? Does the "Speaker Diarization" help when voices overlap? Any advice is appreciated!
Oh, messy Zoom recordings are the worst, but you can salvage a lot! From my own painful experience, uploading the separate audio track rarely helps Otter - their engine seems to process the audio from the MP4 container just fine, and extracting it first can sometimes strip out metadata that helps. Speaker Diarization does help a bit with overlap, but more by isolating chunks to different speakers after the fact than magically untangling the cacophony 😅
My go-to trick is a pre-processing step: run the recording through a free tool like Audacity first. Apply a noise reduction filter (even the basic one helps), then normalize the audio to boost those quiet, far-from-mic voices. You lose the video sync if you do this, but the transcript accuracy jumps noticeably. Then feed that cleaned-up audio file to Otter.
Also, if you have the option on your Zoom account, always record separate audio files for each participant. It's a game-changer for diarization, even if the individual files are low quality.
Pipeline is king.