audio.convert

Re-encode audio to another format at a chosen bitrate. It's the whole audio surface today — one node, two fields, one decode and encode.

Config

Field	Type	Default	Notes
`format`	`mp3` · `m4a` · `opus` · `flac` · `wav`	`mp3`	Output format. Narrower than what can be read — AIFF, WMA, AMR and the rest decode fine as inputs.
`bitrate`	number 8–320 (kbps)	`192`	Ignored by `flac` and `wav`, which are lossless. Out-of-range values are clamped.

Choosing a format

mp3 — plays everywhere. The safe default for anything you hand to a browser or a podcast host.
m4a — AAC. Better than MP3 at the same bitrate.
opus — best-in-class below ~128 kbps. Reach for it on speech and low-bandwidth delivery.
flac — lossless, compressed. Archival, or a master to encode from later.
wav — lossless, uncompressed. Large, and universally readable by tooling.

Each format has exactly one encoder. Vorbis is deliberately absent: asking for it would mean a different codec depending on which machine ran the job.

Cover art travels inside an audio file as a video stream, and carrying it between containers is a common way to fail an encode for a reason that has nothing to do with the audio — so it is dropped. m4a outputs are written with their index at the front (faststart), so a player can start before the whole file arrives.

In a pipeline

{
  id: 'convert',
  type: 'audio.convert',
  config: {
    format: { value: 'opus' },
    bitrate: { value: 96 },
  },
}

With runQueue

tk.runQueue(files, 'audio')
  .convert({ format: 'mp3', bitrate: 192 })
  .options({ poll: { timeoutMs: 600_000 } });   // long files outlast the 120s default wait

There is no audio resize — a branch with no audio.convert passes the source through untouched, which costs an output and a credit for a copy of what you uploaded. Two converts in one branch and the last wins: a format is a single property of the encode.

Every output is a job and a credit, the same as pipeline.output describes.