Prep audio for transcription. The format the service actually wants.
Otter, Notta, Rev, Trint, Descript — they all accept "audio," but each has a sweet spot. Sending a 600 MB stereo WAV when the engine downsamples to 16 kHz mono internally is just slow uploads and wasted quota. Pick your service below, drop your file, get back a file optimized for that service's pipeline.
drop your recording
MP3, WAV, M4A, MP4, MOV, OPUS — anything. Batch supported.
What each service wants, in plain English
- Otter.ai — MP3, M4A, WAV, plus most video. 4-hour cap (paid), 30 min free. Use 96 kbps mono MP3 to upload fast and stay within file-size budgets.
- Notta — Same accepted formats. Free plan: 5 min per recording, 120 min/month. Paid: 90 min per recording. MP3 mono works fine.
- Rev — Human transcription accepts up to 2 GB. AI version is more permissive. Send MP3 128 kbps to balance quality (for human transcribers) and upload speed.
- Trint — MP3/WAV/M4A/MP4. Used heavily in newsrooms. 128 kbps MP3 mono is the standard.
- Descript — Re-encodes everything anyway, so format matters less. Keep stereo if you have multiple speakers panned for clean speaker-detection.
- OpenAI Whisper API — 25 MB per file cap. Engine downsamples to 16 kHz mono internally, so a small MP3 (64 kbps mono) gives the same accuracy as a giant WAV. Use the 64 kbps preset to fit ~50+ min under the cap.
- Google Speech-to-Text — Wants 16-bit PCM WAV at 16 kHz mono for best accuracy. We have a dedicated Whisper / Google STT preset page.
- AWS Transcribe — MP3, MP4, WAV, FLAC, OGG, AMR, WEBM. 2 GB cap. 96 kbps mono MP3 is fine.
Why mono almost always
ASR (automatic speech recognition) engines run on mono. They sum stereo channels internally before processing. Sending stereo costs you 2× upload time and 2× file size for zero accuracy improvement — and sometimes hurts accuracy if the channels are noisy or out of phase.
The one exception: services that do speaker diarization from stereo (separating "speaker 1 in left channel" from "speaker 2 in right channel"). If you recorded an interview with two mics panned hard L/R into one file, keep the stereo. Otherwise, mono.
Why this isn't just "shrink to MP3"
Transcription accuracy is mostly about signal-to-noise ratio, not bit depth. Past about 64 kbps mono MP3, ASR engines hit a quality plateau — more bits don't improve word accuracy. So this tool aims for the smallest file that doesn't sacrifice transcription quality, which is a different optimization than "best-sounding compressed audio."
FAQ
Won't a higher-quality file mean a better transcription?
Up to a point. Past 64 kbps mono, you stop hearing improvements in word accuracy. The engine is the limiting factor, not the audio. So send a small file and save your upload time and storage quota.
What about Zoom recordings — those come as M4A.
M4A is fine to send directly, but Zoom's default M4A is stereo with both speakers in both channels. Force mono here to halve the file size with no transcription quality loss.
Does my audio actually upload through your site?
No. We convert it in your browser, you download the result, then you upload to Otter / Rev / wherever yourself. We never see the audio.
How do I prep audio for OpenAI's Whisper API specifically?
The Whisper preset here gets you under the 25 MB cap. For the local whisper.cpp binary or models that want 16 kHz mono PCM WAV exactly, use our Whisper preset page.
Can it batch multiple files for the same service?
Yes. Drop multiple recordings, pick the service, get a zip of all the prepped files.
What if my source is a video file?
We extract the audio first, then encode in the service's preferred format. Drop the MP4 or MOV directly.
⤳Related tools
Audio for Whisper
Exact 16 kHz mono PCM WAV for whisper.cpp and the Whisper API.
Silence Remover
Trim dead air before transcription — faster, cheaper, more accurate.
Normalize Audio
Even out volume so quiet speakers transcribe as well as loud ones.
Extract Audio
Pull audio out of a Zoom / Loom / Meet video first.