How fast WhipScribe transcribes, measured
Processing time from the production record, 30 days to 27 October 2026, by recording length, format and country. Every figure is computed from the jobs themselves.
The short answer
In the 30 days to 27 October 2026, a one-to-two-hour recording was transcribed in 3 min at the median (2.1 minutes of processing per hour of audio, 407 recordings); a two-to-four-hour recording in 5 min 2 s (1.89 minutes per audio hour). Longer files process faster per hour, not slower.
Measured on production, 30 days to 27 October 2026: 2,617 finished recordings of a minute or more. Processing time is worker time from first word to full transcript; upload and queue time are not included. Quote with the date.
Processing time by recording length
| Recording length | Recordings | Median processing | 90th percentile | Per hour of audio (median) | Audio transcribed |
|---|---|---|---|---|---|
| under 10 minutes | 889 | 13 s | 35 s | 3.6 min | 61 h |
| 10 to 30 minutes | 507 | 50 s | 1 min 54 s | 2.8 min | 161 h |
| 30 to 60 minutes | 438 | 1 min 38 s | 3 min 7 s | 2.3 min | 320 h |
| 1 to 2 hours | 407 | 3 min | 5 min 19 s | 2.1 min | 569 h |
| 2 to 4 hours | 277 | 5 min 2 s | 9 min 32 s | 1.9 min | 728 h |
| 4 to 8 hours | 86 | 3 min 30 s | 16 min 50 s | 0.7 min | 445 h |
| over 8 hours | 13 | 7 min 35 s | 39 min 54 s | 0.9 min | 150 h |
A one-hour file is not a one-hour wait. The median is the typical case; the 90th percentile is what one recording in ten waited longer than.
By format
| Format | Recordings | Median length | Median processing | File size per hour (median) |
|---|---|---|---|---|
| link (YouTube, Drive, podcast…) | 1,104 | 41 min | 1 min 36 s | 58 MB |
| M4A | 511 | 24 min | 1 min 9 s | 58 MB |
| MP3 | 482 | 14 min | 37 s | 58 MB |
| MP4 | 231 | 80 min | 2 min 12 s | 161 MB |
| WAV | 186 | 5 min | 12 s | 115 MB |
| MOV | 20 | 3 min | 31 s | 537 MB |
| OPUS | 20 | 1 min | 7 s | 8 MB |
| OGG | 20 | 4 min | 19 s | 59 MB |
| AAC | 13 | 32 min | 1 min 38 s | 44 MB |
| WEBM | 7 | 5 min | 19 s | 58 MB |
| MPEG | 5 | 4 min | 11 s | 144 MB |
Only the audio is transcribed, so a large video costs the same minutes as its soundtrack. Size per hour is what people actually uploaded, not a bitrate table.
Last 7 days
613 recordings, 620 hours of audio, longest 11.44 hours, largest 1.7 GB, 2.54 minutes of processing per audio hour at the median.
Finished: 2,617 of 2,624 recordings of a minute or more in the window (100%); 7 failed, 0 were still running or were deleted before finishing.
By country
| Country | Recordings | Median processing |
|---|---|---|
| US | 798 | 54 s |
| ? | 243 | 34 s |
| IN | 226 | 42 s |
| PH | 149 | 1 min 12 s |
| DE | 116 | 1 min 49 s |
| IR | 97 | 1 min 11 s |
| FR | 72 | 1 min 15 s |
| GB | 60 | 1 min 36 s |
| CA | 58 | 1 min 11 s |
| BR | 52 | 2 min 24 s |
Processing happens on our own servers, so the country of the uploader does not change the speed; the column is here to show the spread of who uses it.
Questions
How long does WhipScribe take to transcribe one hour of audio?
Measured on production in the 30 days to 27 October 2026: a recording of 1 to 2 hours took 3 min at the median and 5 min 19 s at the 90th percentile, about 2.1 minutes of processing per hour of audio.
Does a longer recording take proportionally longer?
Per hour of audio it is faster: 1.89 minutes per audio hour for 2 to 4 hour recordings against 2.1 for 1 to 2 hours, because the pipeline splits a long file into pieces that run at the same time.
What is the longest recording transcribed recently?
In the last 7 days to 27 October 2026: 11.44 hours in one file, and the largest upload was 1.7 GB. The stated limit is 10 hours or 5 GB per file.
What does processing time measure?
The time from the moment transcription starts on a worker to the moment the full transcript is written, for finished recordings of at least one minute. It does not include the upload itself, which depends on your connection, or any queue wait.
Is this page updated?
Yes. It is regenerated from the production record on every deploy; this version was measured on 27 October 2026 over the previous 30 days (2,617 timed recordings).
Raw data: /proof/turnaround.json. How long recordings really are: /proof/recording-lengths; how much of them is speech: /proof/speech-share. Public long-file runs with their transcripts: /proof/long-files.
Long or large recording? Start at long recordings or large files; the measured limits and prices are on the fact sheet.
See also: Transcription services · Turnaround time · Evidence log · Pricing · Privacy · Terms