Transcribe MPEG

"MPEG" names a standards body, not a file, and three different things people upload carry its name. None of the tools ranking for this search says which is which, so here it is:

The practical answer: it does not matter here. All three upload directly and all three are transcribed the same way. It matters everywhere else — an editor that opens .mp4 may refuse .mpg, and a device that plays .mpg may have never heard of .mp4.

If your file is .mpe, that is a third spelling of the same MPEG-1 container. It is not on our upload list; rename it to .mpg and it will go through, because the bytes are identical.

These are medians measured on our production cluster, expressed as time per hour of audio. A video file adds a little on top while its audio track is extracted.

Scaled to real footage: a 45-minute DVD chapter takes about a minute, a two-hour recorded panel about three minutes, a full six-hour tape capture under nine. The video track costs nothing — it is discarded before transcription, so a 4 GB MPEG-2 file is no slower than the same hour as an MP3.

MPEG files are usually old, and old audio has its own failure modes. None of them stop a transcript, but knowing them tells you where to look when you proofread.

• Tape hiss is constant and broadband. The engine handles it better than it handles crosstalk, but consonants at the end of words are where errors cluster.
• One microphone, whole room. Camcorders record from the camera position, so whoever sits furthest away is transcribed worst. Speaker labels drift on the quiet voices, not the loud ones.
• MPEG Audio Layer II at 128 kbps is the usual audio track on a DVD. That is plenty for speech — the limit on these files is the microphone that recorded them in 1998, not the codec.
• Do not "clean" it first. Noise reduction applied by a video editor removes the same high frequencies the recogniser uses. Upload the original capture.

If the recording has several speakers and matters, turn speaker labelling on: it costs about 2 minutes 7 seconds per hour and turns a wall of text into something you can attribute.

MPEG-2 is the one format where size can bind before length does. DVD-quality video runs roughly 2 GB per hour, so a 5 GB file is only about two and a half hours — nowhere near the 8-hour ceiling.

• Strip the video track. This is the one case where converting first genuinely helps: an hour of DVD video is about 2 GB, and the same hour as audio alone is about 60 MB — the sound is untouched, and the duration limit becomes the only one that matters.
• Split by chapter, not by clock. A ripped disc already has chapter marks at natural boundaries. Cutting there costs you nothing; cutting mid-sentence costs you the sentence.
• Keep the original. Whatever you upload, keep the untouched capture — a transcript can be re-run, a re-encoded tape cannot be un-re-encoded.