Локальный AI-дубляж видео с клонированием голоса

dub-studio is an open-source offline AI video dubbing studio for Windows that re-voices any video into another language — cloning the original speaker's timbre, translating captions, and localizing on-screen text directly on the frame. The build is fully native: Rust + C++/CUDA, a single .exe with no Python or torch; all models and engines download automatically on first launch. Reach for it when you need to translate a training course or YouTube video into another language while keeping the presenter's voice; when footage is sensitive and sending it to the cloud is not an option; when you need auto-generated captions burned into the video; or when you want a diarized transcript split by speaker. Under the hood there are five modes: full dub with voice cloning (Higgs Audio v3, GGUF), voice-over over ducked original audio, subtitles-only, a funny-remix mode that rewrites the whole script, and a diarized transcript export (Sortformer v2, up to 4 speakers). ASR options are Parakeet-TDT (GPU) or Whisper (CPU). Heavy stages — translation, TTS, vision — can optionally be offloaded to the cloud via OpenRouter. NVIDIA GPU is recommended but not required; a CPU-only build is also available. The tool generates dubbed speech output — it is not a transcription-only utility (for that, use a dedicated ASR tool). Works on Windows; macOS and Linux are not supported.