Audio-only cut of a 5 minute episode: mp3 for $0.01

Audio detach turns a finished episode into an mp3 for a flat $0.01 per job, whether the video is 60 seconds or 5 minutes. The source may run to 1800 seconds and the output to 900 seconds. Use it for a listen-only version, a transcript source or a voice-note cut when you already have the video.

TikTok is reported by Metricool to favour videos of 3 to 5 minutes. An audio-only copy gives you a second asset from the same expensive render.

The request

Send the video_url and format. The default is wav (pcm_s16le); mp3 is 128 kbps. channels can be source or mono, and sample_rate can be 16000, 44100 or 48000. A range of start and optional end seconds trims the audio.

curl -X POST https://api.sume.com/v1/audio-detach \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: ep3-audio-mp3" \
  -d '{"video_url":"https://media.sume.com/artifacts/artf_ep3/ep3.mp4","format":"mp3"}'

Limits that matter

The result is a new artifact with audio_url, duration_seconds, format, channels and source_duration_seconds. It is worker ffmpeg, so no model inference.

Check for sound first

A clip with no audio track fails, and a failed job should not surprise you. Run video inspect with frames set to false to read probe.has_audio without making stills. Generated clips from a model with audio off, and silent timeline renders, are the usual cases.

A 5 minute episode at 900 seconds of output headroom is well inside the cap, so no splitting is needed. Longer sources use range to take pieces.

Where it fits

Pair it with captions and a transcript pass for search text, or hand the file to a podcast host. Sume does not publish to feeds or add chapters. It makes the file; where it goes is yours to decide.