YouTube Music: ~1000% faster playback start by getting audio from the player API instead of yt-dlp

[rathlinus]

What does this implement/fix?

Adds a new experimental music provider, YouTube Music (Native).

It talks to the YouTube Music InnerTube API directly instead of going through ytmusicapi, yt-dlp or deno, and it doesn't need a POtoken server. You set it up by pasting the cookie from a logged-in music.youtube.com session.

On the metadata side it covers search, your library (artists, albums, tracks, playlists), album/artist/playlist pages, artist albums and top tracks, similar tracks and recommendations.

For audio it plays the premium 256k streams (itag 141 / 774) by running the player's signature/n cipher in-process with an embedded V8. Without Node, or a browser, no po_token. If that can't run for some reason it falls back to the ANDROID_VR ~150k stream so playback still works. The only new dependency is mini-racer, which ships V8 as a normal pip wheel, so this works in the standard container without anything extra installed.

The existing YouTube Music provider is left as-is; this is a separate one for now.

Tested locally against two Premium accounts: setup, search, browsing an album, and playing a premium track (itag 141, 259 kbps, streams and transcodes fine).

Types of changes

• Bugfix (non-breaking change which fixes an issue) — bugfix
• New feature (non-breaking change which adds functionality) — new-feature
• Enhancement to an existing feature — enhancement
• New music/player/metadata/plugin provider — new-provider
• Breaking change (fix or feature that would cause existing functionality to not work as expected) — breaking-change
• Refactor (no behaviour change) — refactor
• Documentation only — documentation
• Maintenance / chore — maintenance
• CI / workflow change — ci
• Dependencies bump — dependencies

Checklist

• The code change is tested and works locally.
• pre-commit run --all-files passes.
• pytest passes, and tests have been added/updated under tests/ where applicable.
• For changes to shared models, the companion PR in music-assistant/models is linked.
• For changes affecting the UI, the companion PR in music-assistant/frontend is linked.
• I have read and complied with the project's AI Policy for any AI-assisted contributions.
• I have raised a PR against the documentation repository targeting the main or beta branch as appropriate.

[rathlinus]

Click-to-play benchmark

ytmusic_native vs the stock ytmusic provider tested with 12 tracks, on a Premium account

Both play itag141 258 kbps AAC

Step	native	stock	Improvement
Get a playable URL	142 ms	4228 ms	~30× faster
Click to first audio	300 ms	4400 ms	~15× faster

Native spread: 114–452 ms. Stock spread: 3.8–6.1 s.

native does one API call + an in-process cipher solve (~12 ms) against the
cached base.js. Stock spawns yt-dlp every play and solves the challenge with
deno.

One-off: native spends ~1.8 s at startup (grab base.js once), then every play is
fast. Stock pays the full ~4 s on every single play.

per-track (get-URL ms)

videoId	native	stock
-LMuWqr4z5I	452	5032
-eBL6fqrq1s	180	6120
-hqCgw59QXc	125	4210
03ZHI-LYncE	116	4176
0EKyo5HWuXY	114	3812
0X2mn7Sk9lQ	116	4403
0YYWgKTgAQw	118	4198
0f-doEMeI3k	159	4572
0ru6miGzTNs	161	3804
0snTYLgg9w0	115	4246
1DrFDVhb9TY	164	3836
1LPNuFkteI8	169	4765

[rathlinus]

Use this tool to extract the data for the native provider: https://github.com/rathlinus/ytm-login-grabber