feat: dual-layer realtime transcription via RealtimeSTT#9

Open

hoiyada7-maker wants to merge 17 commits into

parkscloud:masterfrom

hoiyada7-maker:realtimeSTT

Commits on Jun 1, 2026

fix: detect GPU via ctranslate2 instead of torch

hoiyada7-maker
and
claude
committed
fix: use nvidia-smi to resolve GPU name and VRAM without torch

hoiyada7-maker
and
claude
committed
fix: gracefully handle missing CUDA runtime DLLs

hoiyada7-maker
and
claude
committed
feat: show dialog when GPU unavailable instead of silent CPU fallback

hoiyada7-maker
and
claude
committed
feat: auto-register NVIDIA pip-package DLL dirs on Windows startup

hoiyada7-maker
and
claude
committed
chore: add nvidia-cublas-cu12 and nvidia-cuda-runtime-cu12 to requirements

hoiyada7-maker
and
claude
committed
fix: prepend nvidia DLL dirs to PATH so ctranslate2 ctypes calls find them

hoiyada7-maker
and
claude
committed
Add custom HuggingFace Whisper models with CTranslate2 conversion

hoiyada7-maker
and
claude
committed
v1.0.4: Variable-length audio chunking with VAD silence detection

hoiyada7-maker
and
claude
committed
feat: hotkey, beep notifications, and clipboard copy on save

hoiyada7-maker
and
claude
committed
fix: add torch to requirements for HuggingFace model conversion

hoiyada7-maker
and
claude
committed
fix: translate all Korean UI strings to English, fix exc scope bug, add transformers dep

hoiyada7-maker
and
claude
committed