AudD for AI startups — AudD resources

If you’re building on audio or media data, you often need to know what recorded music is in a file — to attribute it, tag it, or enrich a dataset with it. AudD is an HTTP API that identifies music against a database of 160 million tracks and returns structured metadata: artist, title, album, label, release date, recording identifiers, and a universal link. This page is the concrete shape of the API and the two common paths: a real-time lookup feature and a bulk enrichment job.

What you can do

Add a “what’s this song” feature

For a short audio clip captured in your product, call the standard endpoint (POST https://api.audd.io/). It returns in under 2 seconds, accepts files up to 10 MB, and returns result: null on no match — distinct from an error, so “no song here” is a first-class answer your UI can render. This is the right path for an interactive, one-clip-at-a-time feature.

Pass a url or upload bytes; the response carries artist, title, album, release_date, label, and song_link (a universal URL on lis.tn).
Use return (Node: returnMetadata, Python: return_metadata) to attach per-provider blocks — apple_music, spotify, deezer, musicbrainz — when your feature links out to a streaming service.

Identify and attribute music in user or training data

For longer audio — full tracks, recordings, video — call the enterprise endpoint (POST https://enterprise.audd.io/). It chunks the file server-side and returns one match per recognized segment, so a single file that contains several tracks yields several attributions. It accepts audio (MP3, WAV, FLAC, M4A, OGG, AAC, WMA, AIFF) and video (MP4, AVI, MOV, MKV, WebM), with no practical file-size cap.

isrc (recording ID) and upc (release ID) come back on enterprise responses for accounts on the Startup plan or higher — the stable identifiers to key a dataset on rather than free-text title matching.
timecode is the position inside the matched track at the recognition point.

Enrich a media dataset at scale

To run an existing archive of files through recognition, the path is a bulk job: iterate your inventory, call the endpoint per item, and write the structured result back to your dataset keyed by your own ID. The enterprise endpoint bills per 12 seconds of audio processed, so cost scales with how much audio you fingerprint — limit and every are the levers that bound it.

Always set limit during development, and on bulk jobs. The enterprise endpoint bills per 12 seconds of audio processed. An unbounded call across a large archive can ingest enormous amounts of audio. Set limit=N per call, and use every=N to sample chunks when you need attribution rather than a complete per-second tracklist.

Read fields beyond the typed surface

Every result and per-provider block exposes an extras map for additional fields the SDK doesn’t surface as typed properties, and every request options struct exposes an extra_parameters map for form fields not exposed as typed parameters (typed parameters win on collision). This is how you read or send fields outside the typed surface without waiting on an SDK release.

Where to start

Scan a bulk audio archive — the enrichment path: iterate an archive, recognize each item, and write structured metadata back to your dataset.
Build a copyright scanner for user-uploaded content — the per-upload recognition pattern, including how to bound enterprise cost on arbitrary user files.
Result fields reference — the exact response schema your dataset writes against, field by field.
Run recognition on Cloudflare Workers — a serverless deployment for the real-time lookup feature.

API teaser

Real-time lookup with the standard endpoint — under 2 seconds, None on no match:

from audd import AudD

audd = AudD("test")  # 10 free requests/day on the public test token

result = audd.recognize(
    "https://audd.tech/example.mp3",
    return_metadata=["apple_music", "spotify"],
)

if result is None:
    print("No song recognized.")
else:
    print(f"{result.artist} — {result.title}")
    print(f"link: {result.song_link}")

Bulk enrichment with the enterprise endpoint — one record per recognized segment, keyed by your own ID:

def enrich(records):
    for rec in records:  # rec = {"id": ..., "url": ...}
        matches = audd.recognize_enterprise(rec["url"], limit=25)
        rec["music"] = [
            {"artist": m.artist, "title": m.title, "isrc": m.isrc,
             "label": m.label, "timecode": m.timecode}
            for m in matches
        ]
        yield rec

The recognition method returns a None/null/nil-equivalent on no match in every SDK. Install for your stack — pip install audd (Python), npm install @audd/sdk (Node), go get github.com/AudDMusic/audd-go (Go) — and see the SDK docs for the other supported languages. Signup includes 300 free requests with no card.

Related