Repository navigation
Lumi v1.6.4
新しくなったこと
- 音声入力がずっと正確に — グラフィックボードのある Windows の PC では、いちばん正確な Whisper のモデル (large-v3-turbo) を GPU で動かして書き起こすようにしました。PowerShell・Git・VS Code のような英単語まじりの言葉も、話し終えてから約1秒で正しく文字になります (初回だけ GPU の準備で少し時間がかかります)
- Whisper のモデルを PC に合わせて選ぶ — 新しい既定
whisper_model autoは、グラフィックボードがあれば turbo、メモリが 8GB 以上なら small、それ以外は base を使います。/install-whisper turboのようにモデルを選んで入れることもできます - 前に base や small を入れていて、グラフィックボードがある PC では、turbo をすすめる案内を出します
whisper_gpu(auto / on / off) で、Whisper で GPU を使うかを選べます。auto は専用のグラフィックボードがあるときだけ使います (内蔵 GPU では CPU より遅いため)
What's new
- Much more accurate voice input — on Windows PCs with a graphics card, speech is now transcribed with Whisper's most accurate model (large-v3-turbo) running on the GPU. Words like PowerShell, Git and VS Code come out right, about 1 second after you stop talking (the very first time takes a little longer while the GPU gets ready)
- Whisper model picked for your PC — the new default
whisper_model autouses turbo with a graphics card, small with 8 GB of memory or more, and base otherwise. You can also choose one, e.g./install-whisper turbo - If you installed base or small before and have a graphics card, Lumi suggests turbo
whisper_gpu(auto / on / off) chooses whether Whisper uses the GPU. auto uses it only with a dedicated graphics card (integrated GPUs were slower than the CPU)
インストール / Install
| OS | |
|---|---|
| Windows 10 / 11 | Lumi-Windows-Setup-1.6.4.exe |
| macOS 11+ | lumi-mac-1.6.4.dmg |
| Linux (x64 / arm64) | lumi_1.6.4_*.deb / lumi-linux-*.tar.gz |
Full Changelog: v1.6.3...v1.6.4