Skip to content

v0.17.2

Choose a tag to compare

View all tags
@steipete steipete released this 11 Jun 21:38
· 198 commits to main since this release
v0.17.2
dad2195

Features

  • CLI media: allow --diarize and speaker identification for local and direct audio/video inputs, including MP3 and MP4, instead of limiting speaker-labelled transcription to YouTube.
  • YouTube transcripts: fall back to Android VR direct audio resolution and configured transcription when yt-dlp is missing or fails, while preserving explicit --youtube yt-dlp and diarization requirements.
  • Chrome extension: transcribe captionless YouTube videos without the daemon using active-player/watch-page or Android VR direct audio, captured SABR fallback, MediaBunny/WebCodecs, and browser-cached multilingual Whisper Tiny.
  • Chrome extension: transcribe fetchable direct and embedded media without the daemon using ranged MediaBunny/WebCodecs decoding, bounded audio chunks, and an idle-evictable browser Whisper runtime.

Fixes

  • Chrome extension: cancel pending summary starts when switching tabs and recover from stalled WebGPU Whisper initialization with a bounded CPU fallback.
  • CLI: use the Codex runtime default model instead of pinning auto fallback to an obsolete model.
  • Daemon: defer cache shutdown until in-flight summary work drains, preventing late writes through finalized SQLite statements.
  • Dependencies: replace Ora, tslog, and the FAL SDK with focused local implementations while retaining spinner, daemon logging, retry, multipart upload, and FAL transcription behavior.
  • Chrome extension: replace the bundled browser FFmpeg WebAssembly runtime with MediaBunny and native WebCodecs, adding AV1 frame extraction while reducing the packaged extension size.
  • Chrome extension: avoid throttled offscreen canvas blob callbacks so MediaBunny slide JPEGs encode in milliseconds instead of roughly one second per frame.
  • YouTube media: prefer direct audio already exposed by the active player or watch page before requesting Android VR media, with resolver and browser decoder diagnostics in extension logs.
  • Extension tests: load the Firefox build through Mozilla web-ext in CI and run daily live YouTube resolver plus daemonless Chrome transcription checks.

Release verification

  • npm CLI: @steipete/summarize@0.17.2, published 2026-06-11T21:35:02.561Z; registry tarball; integrity sha512-A8UnU240wS1REXlfeev1bHmQqmifz3i5qTWqkvobVJMUVDGzdjD3Yxc/xNFHVABnadJFTeITTtWx7NDfdMMCBQ==.
  • npm core: @steipete/summarize-core@0.17.2, published 2026-06-11T21:34:56.578Z; registry tarball; integrity sha512-kSx0wGtx/c2k20dBmTxfsDyFJ0dwEa4Ena2kJDf0cdrDOuoeQCppqrpeaLpt7N7PLYNdMKMvcakeuwLKWkA2GA==.
  • CI proof: PR #261 and Actions run 27378448101 passed core coverage/build/pack, Chromium extension E2E, and Firefox temporary-install smoke.
  • Artifact SHA-256: macOS arm64 c0c65256f36a4359a2dd4bd40dc36719f222cbcb1a244669505226a4c6459c8a; macOS x64 c277c1fc04583c6e7f7cd3d104948e29bbe3f5c406fef7caac74dab2bb85614f; Chrome 4a5cd23830cbc3752562e0a84cbe9aada6d5acfb7334fa6d5612e8b3a9de78f9; Firefox 9e9bbf11e5a49e10da277859966185d6bb6d91679ce1741d29513c4bc6b424cd.
  • Homebrew/core formula is not maintained in this repository; its release autobump typically lands within about three hours.
Assets 6
Loading