Audio and video input
Accepts audio files (wav, mp3, m4a, flac, ogg) and video files (mp4, mov, mkv, avi, webm). For video, the audio is taken from the file before separation.
CatalogLatency Asynchronous; processing time depends on audio duration and the chosen model Async true Rate Limit Per API key Max Input Duration-dependent; longer files take longer
cn8/studio-audio-background-separation
Send an audio file (wav, mp3, m4a, flac) or a video file (mp4, mov, mkv, avi, webm).
Learn more about Audio Background Separation Audio & Voiceasync (job)Instant previewaudioprocessingstudio
Try it
Source type
Audio or video to split
Result
The result shows up here
Upload your audio or video on the left and hit Run.
Overview
Send an audio file (wav, mp3, m4a, flac) or a video file (mp4, mov, mkv, avi, webm). The service removes the vocals and returns the background — drums, bass, and the rest of the instruments mixed into a single track — as a WAV file. The level is balanced so it won't clip. The job runs asynchronously: you get a job_id right away and the background WAV URL when it completes. It's handy for pulling an instrumental for karaoke or remixing, or isolating ambience for analysis.
Key capabilities
Vocal vs. background separation
Vocals are removed and the remaining instruments (drums, bass, and more) are mixed down into a single background track.
Quality vs. speed options
Choose a model that fits your needs: best quality, a faster balanced option, a 6-stem mode that also isolates guitar and piano, or alternatives tuned for music. See the model parameter for the full list.
Clean, balanced output
The background mix is level-balanced so it won't clip. Output is a high-quality stereo WAV.
When to use it
Background music extraction
Pull the instrumental track out of a mixed recording for analysis or remixing.
Speech isolation preprocessing
Use the non-vocal background as a reference for further audio analysis or noise-removal pipelines.
Karaoke and remixing
Remove the vocals from a song to get a backing track for karaoke or arrangement work.
Input & output
input
Audio URL (wav, mp3, m4a, flac, ogg) or video URL (mp4, mov, mkv, avi, webm)
JSON body
output
Background (non-vocal) audio WAV URL, duration in seconds, and usage info
JSON
Guides & tips
How it works
- Submit an audio or video URL. For video, the audio is taken from the file first.
- The vocals are removed and the remaining instruments (drums, bass, and more) are mixed into a single background track, level-balanced so it won't clip.
- The job runs asynchronously: you get a job_id immediately, then poll /v2/jobs/{job_id}. When it completes, result_data carries the background WAV URL — ready to download or use straight away.
Choosing a model
- htdemucs_ft: best quality (default). The slowest option.
- htdemucs: a good balance of quality and speed.
- htdemucs_6s: 6-stem mode that also isolates guitar and piano.
- mdx: fast and tuned for music.
- mdx_extra: slightly better than mdx.
Video input
- Send a video URL. The service takes the audio from it and returns the background as a WAV.
- The video itself isn't modified — only the audio is processed.
Specs
Schema
Request body
mediaUrlrequiredstringPublic URL of the source audio (wav, mp3, m4a, flac, ogg) or video (mp4, mov, mkv, avi, webm)
modelstring · default: htdemucs_ftSeparation model. Options: htdemucs_ft (default, best quality), htdemucs (faster, balanced), htdemucs_6s (6 stems, also isolates guitar and piano), mdx (fast, tuned for music), mdx_extra (slightly better than mdx).
Pricing
Billed per second of background audio output. Two-part tariff: 5 credits base fee per job + 3 credits/second, minimum 5 billed seconds. Failed jobs are not billed.
| Service | Unit | Price |
|---|---|---|
| Audio Background Separation | second | 3 credits/second |
The billed duration is the length of the background audio output.