Skip to content

songid-discovery.md

Latest commit

 

History

History
106 lines (77 loc) · 3.58 KB

songid-discovery.md

File metadata and controls

106 lines (77 loc) · 3.58 KB
Raw

SongID And Discovery

SongID and Discovery Graph form the current music discovery workflow.

SongID

SongID identifies messy music sources and turns them into ranked acquisition paths.

Supported source shapes include:

  • YouTube URLs.
  • Spotify URLs.
  • Direct text queries.
  • Server-side local files.

The evidence model includes MusicBrainz, AcoustID, SongRec, transcripts, OCR, comments, chapters, provenance, perturbation probes, Panako, Audfprint, Demucs stems, C2PA hints, and forensic lanes.

SongID runs in a durable queue with persisted queue position and worker slot. The configured concurrency is controlled by songid.max_concurrent_runs / --songid-max-concurrent-runs / SONGID_MAX_CONCURRENT_RUNS.

Runtime provider availability is exposed at:

GET /api/v0/songid/capabilities

The capability response is the operator-facing truth for optional providers such as yt-dlp, ffmpeg, SongRec, Panako, Audfprint, Demucs, Whisper, Tesseract, C2PA, MusicBrainz, AcoustID, and Chromaprint. The intentionally unsupported HashFromAudioFileEnabled hardening flag is reported as broken and unavailable.

Docker images include the cheap core prerequisites by default. For heavyweight experimental lanes, run the container installer as an operator:

docker exec -u root slskd install-optional-media-tools distro ai-python c2pa audfprint

Panako requires a deliberate jar source:

docker exec -u root \
  -e PANAKO_JAR_URL=https://example.invalid/panako.jar \
  slskd install-optional-media-tools panako

The UI does not auto-install these tools when a feature is enabled because the toolchains change the container supply chain and require elevated install permissions. The capabilities response names the missing command or file and, for Docker, points at the matching installer profile.

Result Review

SongID surfaces:

  • Identity and synthetic assessments.
  • Top evidence for and against a result.
  • Quality class and perturbation stability.
  • Known-family and generator-family context.
  • Forensic matrix export/debug views.
  • Split track plans for mixes and long uploads.
  • Candidate fan-out actions for ambiguous or multi-segment sources.

Synthetic or AI-origin signals are contextual. They should not override strong catalog identity when ordering download actions.

Manual Search stays direct: typing a query and pressing Search opens results without asking the user to approve the same query first.

Discovery Graph

Discovery Graph / Constellation provides a typed graph of nearby tracks, albums, artists, SongID runs, MusicBrainz targets, and search-result seeds.

Current surfaces include:

  • SongID mini-map and modal.
  • MusicBrainz graph launchers.
  • Search list/detail graph launchers.
  • Search-result graph glyphs.
  • In-page atlas panel.
  • Dedicated /discovery-graph route.

Current actions include recenter, queue nearby, pin, compare, save branch, copy branch reports, and export graph/evidence context.

Remaining Research Scope

The remaining SongID/Discovery work is research/depth work, not missing baseline UI:

  • Better mix detection for long gaps, overlaps, chapters, and comments.
  • Deeper in-panel SongID queue/perturbation/fan-out UX.
  • More Discovery Graph provenance/evidence drill-down and semantic zoom.
  • Essentia-backed MIR where it materially improves identification.
  • Melody or cover-similarity matching.
  • Embedding-based local family clustering.

These items should stay review-first and network-conscious. They should not start peer searches, browse peers, queue downloads, or mutate files unless the user explicitly triggers an action with visible impact.