This release adds 3 notable features for engineering teams evaluating rollout.
✓ No known CVEs patched in this version
Topics
+1 more
Summary
AI summaryAdded CLAP audio embeddings support for detection, embedding generation, and similarity search.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Feature | Medium |
Adds audio embedding capability via CLAP backend. Adds audio embedding capability via CLAP backend. Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Feature | Medium |
Introduces `build_embeddings(kinds=["audio"])` to generate CLAP audio embeddings. Introduces `build_embeddings(kinds=["audio"])` to generate CLAP audio embeddings. Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Feature | Medium |
Adds `find_similar(kind="audio")` for querying audio embeddings. Adds `find_similar(kind="audio")` for querying audio embeddings. Source: llm_adapter@2026-06-12 Confidence: low |
— |
| Feature | Low |
Adds 7 new unit tests covering audio backend detection, idempotency, and graceful skip behavior. Adds 7 new unit tests covering audio backend detection, idempotency, and graceful skip behavior. Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Feature | Low |
Adds `find_similar(kind="audio")` for querying audio vectors and free-text via CLAP text encoder. Adds `find_similar(kind="audio")` for querying audio vectors and free-text via CLAP text encoder. Source: granite4.1:30b@2026-06-12-audit Confidence: low |
— |
| Bugfix | Low |
Gracefully skips audio processing when CLAP backend is missing, showing install guidance. Gracefully skips audio processing when CLAP backend is missing, showing install guidance. Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Refactor | Low |
Updates `capabilities` to include an `audio` block with install guidance and Tools-page entry. Updates `capabilities` to include an `audio` block with install guidance and Tools-page entry. Source: granite4.1:30b@2026-06-12-audit Confidence: low |
— |
Full changelog
CLAP audio embeddings — the final phase of the post-program improvements (v2.46.1 → v2.51.0). Detection-based like every other backend (never auto-installed); local compute, so nothing touches the caps ledger.
Added
- Audio backend detection: CLAP via
transformers(laion/clap-htsat-unfused, preferred) or thelaion_clappackage, plus torch + ffmpeg.capabilitiesgains anaudioblock with install guidance and aclap_audioTools-page entry. Activate withpip install transformers(model weights ~600 MB download on first use). build_embeddings(kinds=["audio"]): one CLAP window per shot (center-cropped to ~10 s, piped from the source media as raw PCM — read-only on source media, no temp files) plus a clip-level mean vector, stored asembedding_kind="audio"rows. Idempotent via content hashes; offline media is reported inskipped_missing_media; a missing backend is a graceful skip with install guidance.find_similar(kind="audio"): shot/clip queries over the audio vectors, and free-text queries ("engine revving") via the CLAP text encoder.
Validation
- Backend mocked in 7 new unit tests (build plumbing, idempotency, clip mean vector, graceful skip, missing media, audio similarity, kind validation); offline suite at 1184.
- Graceful-skip path verified live on a machine without a CLAP backend: clean errors with install guidance from
build_embeddings,find_similar, and the capabilities/Tools surface.
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
Related context
Beta — feedback welcome: [email protected]