Skip to content

LinguaRelay v0.1.0

Choose a tag to compare

View all tags
@MuzeAnisichael MuzeAnisichael released this 11 Aug 07:09
· 17 commits to main since this release
v0.1.0
2633046

LinguaRelay v0.1.0

This is the first public Windows x64 release of LinguaRelay. It captures computer output audio in the background, recognizes a manually selected source language, translates locally between Chinese, Japanese, English, and Korean, and renders either translated-only or bilingual captions.

Download

  • LinguaRelay-0.1.0-Setup-x64.exe: per-user Windows installer (recommended).
  • LinguaRelay-0.1.0-Windows-x64-portable.zip: portable application directory.
  • LinguaRelay-0.1.0-models.zip: optional manual download of the pinned local models; the app can download it on first launch.
  • LinguaRelay-0.1.0.spdx.json: SPDX 2.3 software bill of materials.
  • SHA256SUMS.txt: SHA-256 checksums for every release asset.

On first launch, review the model licenses and allow the approximately 1.36 GiB installed model pack. Audio recognition and fast translation then run locally. LLM correction remains off until explicitly configured.

Verification and security notice

Verify the downloaded asset against SHA256SUMS.txt. These first binaries are not Authenticode-signed, so Windows SmartScreen may show an unknown-publisher warning. Checksums protect against accidental corruption when obtained from the same GitHub release, but do not provide an independent publisher identity.

See docs/PRIVACY.zh-CN.md, docs/THREAT_MODEL.md, SECURITY.md, and THIRD_PARTY.md before using optional cloud correction or private content.

Measured release gate

  • 95 unit/integration tests passed; Ruff passed.
  • 20 legal four-language acoustic-condition samples; weighted WER/CER 21.13%.
  • All 12 translation routes measured; mean chrF++ 38.573 on the small CC0 corpus.
  • Protected terminology retention: 48/48 (100%).
  • CPU end-to-end: first caption 2.98 s P50 / 11.77 s P95; final pipeline RTF 0.189.
  • M4 non-blocking/disconnect/circuit/revision fault gates passed.

The corpus uses pinned FLEURS speech under CC-BY-4.0 and deterministic project-generated acoustic simulations. Meeting/video/music labels describe signal conditions, not real meeting or copyrighted entertainment content.

Assets 7
Loading