Keep this tab open — you'll be redirected to your transcript.
Field recordersStudio sessions24-bit / 96 kHzDictaphones
WAV is a size problem, not a quality problem
WAV is uncompressed PCM, and that's the one thing to plan around: CD-quality stereo runs about 10 MB per minute, so the 100 MB anonymous cap fits roughly 9-10 minutes of stereo (about 19 mono) — you'll hit the size limit long before the 30-minute one. A free account raises the cap to 500 MB, paid tiers to 5 GB.
WAV is also the one format where converting before upload genuinely helps. The model listens at 16 kHz mono internally, so a command like "ffmpeg -i in.wav -ar 16000 -ac 1 out.flac" shrinks the file about tenfold while staying lossless for speech purposes — the transcript comes out identical.
From studio and field recordings to text
Everything a WAV workflow produces decodes here: 16- or 24-bit, 44.1 to 192 kHz, broadcast WAV from field recorders, and the ADPCM or µ-law variants old dictaphones and phone systems wrote. High resolution doesn't add accuracy — the audio is downmixed and resampled internally — but it never hurts either.
Transcript timestamps always start at zero: embedded BWF timecode is ignored. If the WAV belongs to a video edit, generate the SRT or VTT here and apply your editor's offset when you import it.
Frequently asked questions
Will an uncompressed WAV transcribe more accurately than a compressed copy?
Not measurably. Accuracy is set by the recording itself — mic distance, noise, crosstalk — long before the codec matters, and the model downsamples everything to 16 kHz mono anyway. Upload the WAV if that's what you have, but don't hunt for a lossless master expecting a better transcript.
My WAV hits the 100 MB limit before 30 minutes — why?
Uncompressed math: CD-quality stereo is about 10 MB per minute, so 100 MB holds only 9-10 minutes. Three ways out: a free account raises the cap to 500 MB, downmixing to mono halves the size, or convert to FLAC at 16 kHz mono — roughly 1 MB per minute and losslessly identical to the model's ears.
Are 24-bit and 96 kHz studio WAVs supported?
Yes — every bit depth and sample rate FFmpeg reads: 16/24/32-bit, 8 to 192 kHz. They're resampled internally, so the extra resolution neither helps nor hurts the transcript; it just makes the upload larger. Bounce a 16-bit/44.1 kHz or mono copy if size is the bottleneck.
How are stereo and multi-channel WAVs handled?
Channels are downmixed to mono before transcription. That's fine when every channel carries the same room — but if your recorder put different people on separate channels (lav left, boom right), split the channels and transcribe the files separately; you'll get cleaner text per person than any downmix.
Does broadcast WAV (BWF) timecode carry into the transcript?
No — embedded BWF timecode and cue markers are ignored, and the transcript's timestamps start at 0:00 from the top of the file. If the recording belongs to a synced video edit, export SRT or VTT here and let your editing software apply the offset on import.
Old dictaphone and telephony WAVs (ADPCM, µ-law) — supported?
Yes. FFmpeg decodes the legacy WAV variants — ADPCM, µ-law/A-law, GSM — that dictation hardware and phone systems wrote. Expect more errors than a modern recording: those formats are 8 kHz narrowband, so names and quiet passages need a pass in the editor. Still beats retyping a tape.