Repository navigation
v0.1.1 — Return to the main page
Fixes getting stuck in Buzz’s live-transcription window when no microphone is configured.
- Adds Back to main page beside Upload files and Download transcripts in the browser’s top bar.
- Closes the live-recording window through Buzz’s normal cleanup and brings the existing file list forward.
- Keeps the application, saved history, and file-transcription jobs in the same running session.
- Keeps the toolbar usable on narrow screens.
Install or update
Versioned image: ghcr.io/stormavenger/buzz-unraid:0.1.1
Stable update channel: ghcr.io/stormavenger/buzz-unraid:latest
Use the current Buzz-Transcription Unraid template. New installs use /mnt/user/appdata/buzz-transcription; existing installations should retain their saved folder mappings. Back up appdata before upgrading.
The publication workflow passed. Both tags identify sha256:4b5920b92d4b1b1ba770f5916a5a96ef2716864d6d9857fa4aa454e0a5b5b0af. Matching sources, notices, inventory, scan reports, and checksums are attached below.
Validation
The Linux amd64 build and browser/startup/persistent-storage checks passed. The real Qt test opens the live window without a microphone, returns through the same shortcut sent by the toolbar, repeats the operation, and verifies the existing main window and database are retained. Chrome testing also returned from the live view using the new top-bar button.
The release scan found zero HIGH or CRITICAL findings; 645 MEDIUM and 32 LOW findings remain visible in image-scan.json. Matching-source collection completed with 1,064 archives and no failures.
Scope
Browser microphone forwarding is not configured by this release. This remains a single-user desktop for a trusted LAN or VPN. GPU requirements and the Pascal-compatible CUDA 12.6/PyTorch 2.8 stack are unchanged.
Container-specific code and original assets use MIT. Bundled software retains its own licenses; see the notices and corresponding-source records attached to this release. Model weights are downloaded separately.
Archived validation details
Moved from docs/VALIDATION.md on October 28, 2026. These are historical results; stable-tag statements describe the time of the original checks.
Release commit 3dd353e8b0afe180e0daab72e8daca8cc7d6089f passed the Linux build and browser/persistence check. The build exercises the real live-recording window with no microphone, uses the application shortcut from text-field focus, repeats the return, and verifies the same main-window and database objects remain open. The 16 host tests and JavaScript syntax check also passed.
Chrome testing opened Live Recording and returned to the main file list using Back to main page in the top bar, including repeated use. Testing used a disposable desktop with fresh application data. No personal recordings were used, and this release does not add browser microphone forwarding.
The v0.1.1 publication workflow repeats the image build, runtime checks, clean-image inventory, scan gate, and matching-source collection before publication. The CUDA/PyTorch dependency versions are unchanged.
The v0.1.1 release scan reports zero HIGH or CRITICAL findings, with 645 MEDIUM and 32 LOW findings retained in the report. Matching-source collection completed with 1,064 archives and no failures. The image labels identify version 0.1.1 and the tested commit above. Reports and sources are attached to the matching release.
Published v0.1.1 and stable :latest image digest: sha256:4b5920b92d4b1b1ba770f5916a5a96ef2716864d6d9857fa4aa454e0a5b5b0af.
An anonymous pull with an empty Docker credential configuration succeeded on Unraid and matched that digest. The exact published image passed CUDA computation and offline Faster Whisper Small transcription of a public speech sample on the GTX 1080. A separate disposable container became healthy, reported version 0.1.1 and the release commit, and passed the Chrome live-recording-to-main-page button check with fresh application data. Both test containers were removed after verification. The production application was not reinstalled; installation is left to the user.