Transcribe OGG

Voice messages from WhatsApp, Signal and Telegram are Ogg Opus. They arrive as .opus or .ogg depending on which app exported them, and both upload here unchanged — no converting to MP3 first, which would cost a generation of audio quality before the transcriber ever saw it.

Getting the file out of WhatsApp

1. On a phone: long-press the voice message → Share → save to Files, Drive or email it to yourself. The saved file is .opus.
2. On WhatsApp Web or Desktop: hover the message → the chevron → Download. You get .ogg.
3. A whole chat at once: chat menu → Export chat → Attach media. The archive contains every voice note as a separate file.

Drop any of them into the box at the top of this page. A two-minute voice note is transcribed in roughly three seconds, and the first five minutes need no account.

Voice notes are recorded at low bitrates — often 24–32 kbps — which is exactly the range Opus was designed for and exactly where MP3 starts smearing consonants. That is why a voice message that sounds thin still comes back readable.

These are medians measured on our production cluster, converted to time per hour of audio. They are not estimates, and they are not a best case recorded on a quiet studio file.

Scaled to the files people actually upload: a two-minute voice note takes about three seconds, a 40-minute meeting about a minute, a three-hour recording about four and a half minutes. Queue time at peak hours sits on top of all of these and has nothing to do with the model.

Not for uploading: both are read directly, and so is the .oga extension. It matters for what you can expect back, because the two codecs fail differently as the bitrate drops.

If you have a choice when exporting — from Audacity, OBS or a recorder app — Opus at 32 kbps produces a smaller file that transcribes as well as Vorbis at three times the size. If you are handed a file, upload it as-is: re-encoding to "improve" it only adds a second round of loss.

Eight hours is the ceiling on the Premium plan, and it binds long before the 5 GB size limit does — an hour of OGG at 128 kbps is only about 58 MB, so a 5 GB file would be somewhere near 90 hours of audio. In practice length is what stops an upload, not size.

• Split at silence, not at the clock. Cut on a pause between sentences and keep a few seconds of overlap; a cut mid-word costs you the word on both sides.
• Split by session, if there is one. A day-long conference is really six talks — six uploads give you six transcripts you can actually navigate, and speaker labels restart cleanly per file.
• Do not re-encode to shrink it. Length is the limit, so a smaller file at the same duration buys nothing and costs quality.