Voiceover spine: wav or mp3? 1,800 s is 158.76 MB or 28.8 MB

Keep a voiceover spine as wav while you are still joining and editing it, and make mp3 only for the last copy. Sume's Timeline audio returns wav (pcm_s16le, sample-exact) by default or mp3 (smaller, but it adds priming padding again at every edge), so a 20-part mp3 concat has 19 seams that each carry padding. At the 1,800-second maximum a 44.1 kHz mono wav is about 158.76 MB and a 128 kbps mp3 is about 28.8 MB.

Sizes at 1,800 seconds

These are arithmetic from the sample formats in the docs (16-bit samples, the listed sample rates, and the 128 kbps rate the audio detach docs give for mp3). Real files add a small header, and mp3 sizes vary slightly with the encoder, so treat them as estimates.

Which seams matter

The Timeline audio docs say to keep wav if the file will be joined again or if it drives lip-sync, because each mp3 edge adds priming padding. A concat of 20 parts, with the 20-part maximum, makes 19 internal joins. If those parts are mp3, the padding adds up across the joins; if they are wav, the join is gapless in the sample domain.

Joins cost $0.01 per job whatever the format, so the choice is about quality and storage, not price.

A practical rule

Generate and join in wav. Detach audio from video in wav at 16 kHz mono when the only reader is STT (32,000 bytes a second, 19.2 MB per 600 seconds). Export mp3 once, from the finished spine, when file size matters for delivery.

Storage over a catalog

Sizes add up. A library of 200 spines at the 1,800-second maximum is 200 x 158.76 MB = 31.75 GB as 44.1 kHz mono wav and 200 x 28.8 MB = 5.76 GB as 128 kbps mp3, a ratio of 5.5 to 1. Most spines are far shorter than the maximum, so scale by your average length. If storage matters, keep wav for the spines still in production and re-export the finished ones.

The size is also a transfer time. At 25 MB a second the 158.76 MB file takes about 6.4 seconds to move and the mp3 about 1.2 seconds, which matters little for a render but adds up in a loop over many files. These two transfer times are arithmetic at an assumed link speed, not measurements.

The practical advice is short: keep wav while you are still editing seams, because it is sample-exact, and choose mp3 only for the final delivery copy where size matters more than seam accuracy. Because mp3 frames add padding at each edge, 19 joins in a 20-part chain accumulate 19 small offsets that wav does not have.