audio_spine_low_fidelity: why a 16 kHz transcript wav warns

audio_spine_low_fidelity is a warning Timeline 1.0 raises when the render's audio.url is a speech-to-text file, a 16 kHz mono wav, or any spine under 32 kHz, or mono under a stereo video source. Detach the track with audio_detach at the source sample rate and channels, and use that as the spine.

Where the file comes from

Video inspect returns a transcript whose audio_url is a 16 kHz mono pcm_s16le wav. The API schema calls it speech-to-text input, reusable as an STT audio_url or a voice-clone sample, and says it is not a timeline spine, because everything above 8 kHz and the stereo image are gone.

The fix

Call POST /v1/audio-detach with format: "wav", leave channels and sample_rate unset so they inherit the source, and pass the returned artifact URL as audio.url on the render. It costs $0.01 per job. The master inherits the spine's rate and channels, so a poor spine produces a poor render.

It is a warning, not a failure

The render still runs, and the master can only carry what the thin spine kept. Check the job's warnings before you ship. The source channels post explains the inherit behaviour in detail.