Convert audio to MIDI - AudioPod AI

What you get: a .mid file per transcribed part (bass, vocals, piano by default; guitar on request) and one multitrack .mid with every part, plus the detected tempo. Input: a local file or a URL of a recording at least 6 seconds long. A full mix is fine: AudioPod separates the parts first.

1. Install and set your key

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

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

2. Convert

from audiopod import Client

client = Client()

# convert() waits for the job by default
job = client.midi.convert(
    file="song.mp3",  # or url="https://example.com/song.mp3"
    stems=["bass", "vocals", "piano"],
)

print("multitrack:", job["merged_midi_url"])
for part, link in (job.get("midi_urls") or {}).items():
    print(part, link)

import AudioPod from "audiopod";

const client = new AudioPod();

// transcribe() submits the conversion and waits for it
const job = await client.midi.transcribe({
  file: "./song.mp3", // or url: "https://example.com/song.mp3"
  stems: ["bass", "vocals", "piano"],
});

console.log("multitrack:", job.merged_midi_url);
for (const [part, link] of Object.entries(job.midi_urls ?? {})) {
  console.log(part, link);
}

Already separated the stems?

Transcribe a finished stem separation job instead of separating again. It is billed at the lower add-on rate.

midi = client.midi.convert_from_stem_job(stem_job_id=stem_job["id"])  # waits by default

const queued = await client.midi.convertFromStemJob(stemJob.id); // returns at once
const midi = await client.midi.waitForCompletion(queued.id);

Know before you ship

It is a starting-point transcription: expect to tidy note timing and lengths in your DAW. Dynamics are approximate.
Drums are not transcribed. Requesting drums or other returns them in skipped_stems.
Guitar is experimental and can over-detect notes on busy mixes, so it is off unless you ask for it.
Dedicated piano and vocal engines need a Creator plan or above. See pricing.

From an AI agent

With the AudioPod MCP server connected (core or full): “Turn the bass line in https://example.com/song.mp3 into MIDI.” The agent calls audio_to_midi with stems: ["bass"]. Full endpoint reference: Audio to MIDI.