Transcribe MKV

Matroska exists to hold everything at once: several video tracks, several audio tracks, subtitles, chapters, attachments. That flexibility is what makes it the container of choice for archived films and recorded streams.

Here, the first audio track is the one transcribed. If your file carries a dub, a commentary and an original language track, the first one wins — and if that is not the one you want, extract it before uploading: ffmpeg -i in.mkv -map 0:a:1 -c copy track2.mka pulls the second audio track without re-encoding.

Existing subtitle tracks are not read. The transcript is produced from the audio, so it will not inherit errors or timings from a subtitle already in the file.

These are medians measured on our production cluster, expressed as time per hour of audio.

Against the MKV files uploaded here: the median runs 20 minutes 57 seconds and comes back in about 31 seconds. Nine in ten are under 47 minutes. The video track costs a little while the audio is extracted, and nothing after that.

The median MKV uploaded here is 181 MB against 47 MB for MP4, and the reason is bitrate: 2,611 kbps against 684. Matroska is the container people use when they are not compressing aggressively — archived Blu-ray rips, high-bitrate recordings, multi-track masters.

None of that reaches the transcriber. The audio is extracted and the rest discarded, so a 5 GB MKV and a 500 MB MP4 of the same length take the same time.

Size, and sooner than you would expect. At 2,611 kbps the 5 GB ceiling arrives at roughly four hours, well before the 8-hour limit — and the largest MKV processed here was 5.10 GB, right at the boundary.

• Extract the audio if the file is near the limit. ffmpeg -i in.mkv -map 0:a:0 -c copy audio.mka takes seconds and removes the problem.
• Split at chapter marks. An archived recording usually has them, and they fall at natural boundaries.