Drop a recording, a voice note or a video. The page reads its own header and tells you the real format, the codec, the length and the bitrate, then what transcribing it would take. The file stays on your device.
Bitrates read from files transcribed on WhipScribe, from April to October 2026. They are why the same hour of speech can be 6 MB or 600 MB.
| File | Where it comes from | Typical bitrate | One hour is about |
|---|---|---|---|
| .opus | WhatsApp voice notes (PTT-…-WA….opus on Android, …-AUDIO-….opus in an exported chat) | 16 to 19 kbps | 7 to 9 MB |
| .ogg | Voice notes saved from WhatsApp Web (WhatsApp Ptt ….ogg) and Telegram | 16 to 19 kbps; Telegram up to 31 | 7 to 14 MB |
| .webm | Browser and Chrome-extension recordings, including WhipScribe's own recorder | 129 kbps | 58 MB |
| .gsm | Phone-system prompts and voicemail (Asterisk, FreePBX) | 13.2 kbps | 6 MB |
| .m4a, .mp3 | Phone voice memos, podcasts, recorder exports | median 58 MB per hour | 58 MB |
| .wav | Field recorders, DAWs, dictation exports | median 115 MB per hour; 16-bit 44.1 kHz stereo is 1,411 kbps | 115 to 635 MB |
| .mp4 | Phones, cameras, meeting recordings | median 192 MB per hour | 192 MB |
The m4a, mp3, wav and mp4 medians are from 30 days of production to 2026-10-16 (449, 402, 265 and 279 recordings); the voice-note rows are per-file bitrates of the uploads themselves. None of it changes what transcribing costs: you pay by minutes of audio, not by megabytes. The full measured breakdown, links and devices included, is on what people bring to transcribe, and how long those recordings run is on real recording lengths.
Every format begins with a signature: RIFF…WAVE for WAV, OggS for Ogg, ftyp for MP4, M4A and MOV, 1A 45 DF A3 for WebM and MKV, #!AMR for AMR, \x02dss and \x03ds2 for Olympus and Philips dictation. After the signature the header holds the codec, the sample rate, the channels and, in most formats, the exact length. Where a format has no length in its header (raw AAC, constant-bitrate MP3, browser WebM), the inspector works it out from the bitrate or asks your browser to scan the file, and says which it did.
It also catches the cases where the name lies: a WeChat voice message saved as .amr that is really SILK, an .m4a that is an Audible book, a .mov with no sound track, a video with two audio tracks. For each format there is a page on where it comes from and how to export it: Opus and WhatsApp voice notes, WebM, AMR, DS2 dictation, WMA, MKV and every other format.
Look at its first bytes, not its name. Every format starts with a signature: RIFF and WAVE for WAV, OggS for Ogg (WhatsApp and Telegram voice notes), ftyp for MP4 and M4A, 1A 45 DF A3 for WebM and MKV, #!AMR for AMR. This page reads those bytes in your browser and names the container and the codec inside it.
Recordings made in a browser or by a Chrome extension are written as they are captured, so the file never gets a length in its header. Players cannot show a length or seek, but the audio is all there and transcribes normally. The inspector asks your browser to scan the file for its length instead.
On WhipScribe's production servers, the median over 30 days to 2026-10-16 was 3.38 minutes of processing per audio hour for recordings under 10 minutes, 2.15 for recordings of 1 to 2 hours and 1.74 for recordings of 8 hours and longer, from 2,241 timed recordings. The inspector applies the right band to your file's length.
No. The page reads the first and last megabyte of the file in your browser to find its header, and nothing leaves your device. Only if you then choose to transcribe it is the file uploaded.
Choose the same file here for a timestamped transcript. The opening minutes are a free preview.
Once the file is ready, send it as it is: see MP3 to text for long recordings, the MP3 to text converter and how to transcribe a podcast.