Separate a song into stems - AudioPod AI

What you get: one audio file per stem, as download links keyed by stem name (vocals, drums, bass, other, …). Input: a local file or a URL of an audio or video file.

1. Install and set your key

pip install audiopod
export AUDIOPOD_API_KEY="ap_..."

npm install audiopod
export AUDIOPOD_API_KEY="ap_..."

2. Separate

from audiopod import Client

client = Client()

# separate() waits for the job by default
job = client.stem_extraction.separate(
    file="song.mp3",  # or url="https://example.com/song.mp3"
    mode="four",      # vocals, drums, bass, other
)

for stem, link in job["download_urls"].items():
    print(stem, link)

import AudioPod from "audiopod";

const client = new AudioPod();

// separate() submits the job and waits for it
const job = await client.stems.separate({
  file: "./song.mp3", // or url: "https://example.com/song.mp3"
  mode: "four",
});

for (const [stem, link] of Object.entries(job.download_urls ?? {})) {
  console.log(stem, link);
}

Modes

`mode`	Stems	Plan
`two`	vocals + instrumental (karaoke / acapella)	every plan
`four`	vocals, drums, bass, other	every plan
`six`	vocals, drums, bass, guitar, piano, other	every plan
`single`	one core stem (set `stem`, e.g. `"vocals"`) plus everything else	every plan
`producer`	up to 8, with the drum kit split into kick, snare, hi-hat	Creator and above
`studio`, `mastering`	up to 12 and up to 16	Pro and above

The detailed modes return “up to” their count: a stem the track doesn’t contain isn’t returned. To isolate one specific instrument (for example saxophone or violin), pick it from the 45-instrument catalog with the advanced endpoint; see Stem Separation. See pricing for plans.

Tips

Karaoke track and acapella: mode="two" returns both. For one core stem plus everything else, use mode="single" with stem set to vocals, drums, bass, guitar, piano or other.
Download links are short-lived. Fetch a fresh set with client.stem_extraction.get_job(id) (Python) or client.stems.get(id) (Node).
Feed a finished stem job straight into audio to MIDI to transcribe its parts without separating again.

From an AI agent

With the AudioPod MCP server connected: “Split https://example.com/song.mp3 into 6 stems.” The agent calls separate_stems with mode: "6stem". Over MCP the modes are 4stem, 6stem, 2stem_vocals and 2stem_other. Full endpoint reference: Stem Separation.