Timeline audio mp3 adds padding at every edge: keep wav to rejoin

Use wav when you will join audio again. Sume timeline audio defaults to sample-exact wav (pcm_s16le), while mp3 is smaller but adds priming padding at every edge, so parts cut and re-joined as mp3 pick up a small gap at each seam. The docs say to keep wav if the file will be joined again or if it drives lip-sync.

The job is POST /v1/timeline-1.0/audio, $0.01 per job, per the timeline audio docs. It runs worker ffmpeg only, with no provider inference.

Pick the format by what happens next

operation: concat joins 1 to 20 parts[]. operation: split cuts one file into 1 to 20 ranges[]. Each result is a new durable artifact. The output.format field is wav or mp3.

A split-then-join example

Say a 60-second spine is split into three 20-second ranges, then the middle one is replaced and the three are joined. With wav that is two jobs, 2 x $0.01 = $0.02, and the join has no added gaps. With mp3 the same two jobs add padding at the two seams you created plus the edges, which you will hear as a small hitch.

If you need to cut many ranges from one video track, detach the audio once with audio detach ($0.01), then split that file. Detaching per range repeats the extraction and the fee.

{
  "operation": "split",
  "url": "https://media.sume.com/artifacts/artf_demo/spine.wav",
  "ranges": [
    { "start": 0, "end": 20 },
    { "start": 20, "end": 40 },
    { "start": 40, "end": 60 }
  ],
  "output": { "format": "wav" }
}

Practical rules

Source files must already live on your workspace's media host, so import first with POST /v1/media-imports. Idempotency-Key is required on writes, and the result of each job is read from the job envelope. Parts and ranges can be trimmed with source_in and duration fields per part, so a single concat job can drop a breath or a click at the start of a line without a separate step.

• Detach once ($0.01), split once ($0.01), join once ($0.01): $0.03 for a full edit pass.
• Convert to mp3 only as the last step, after every join.
• A part list or range list is limited to 20 entries per job, so a longer edit needs two jobs.

Why the padding happens

The mp3 format works in fixed-size frames, and encoders add a short run of silence at the start and sometimes the end so decoding can warm up. When a file is cut at a time that does not match a frame boundary, that padding appears in each piece. One seam may be hard to hear, but a join of twenty parts repeats the padding twenty times, and the docs say it appears at every edge.

The practical conclusion matches the docs: keep the working copies as wav, which is sample-exact, and encode a single mp3 at the end if file size matters.