Trim a voiceover take with Timeline audio: source_in and duration

Send a concat request with a single part and set source_in (where to start, in seconds) and duration (how long to take). Sume returns a new gapless wav of just that slice for a flat $0.01 per job. The audio must already be hosted on Sume, such as a TTS result.

The request needs an Idempotency-Key header, so a retry returns the same job instead of a second charge.

The request

Set SUME_API_KEY and TAKE_URL (a Sume-hosted audio URL) in your shell, then run it.

curl -sS https://api.sume.com/v1/timeline-1.0/audio \
  -H "x-api-key: $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: trim-take-4-v1" \
  -d '{
    "operation": "concat",
    "parts": [
      { "url": "'"$TAKE_URL"'", "source_in": 1.2, "duration": 8.5 }
    ],
    "output": { "format": "wav" }
  }'

Limits to know

The contract and docs set these bounds.

Why wav is the default

The reference explains that wav stays sample-exact, so the result can be joined again without picking up encoder delay, while mp3 adds priming padding at every edge. If you plan to trim, then join, then trim again, keep every intermediate in wav and convert to mp3 only on the last pass.

Cut and join in one call

Each part carries its own source_in and duration, so one call can pull the best sentence from three takes and join them in order. For ranges out of one long file, use operation: split with a url and up to 20 ranges, which return one file per range. That is a good fit for pulling clips out of a long TTS read.

The route works only on audio you already have on Sume. For the whole-video case, a Timeline 1.0 render can take the audio parts directly, so you can skip this job when the audio is only needed in one render.