Download VoiceStudio
VoiceStudio runs voice cloning, dubbing, and audiobook production on local hardware without accounts or subscriptions. The program includes a model catalogue for TTS, ASR, and LLM engines, a batch queue for large jobs, dictation with system-wide shortcuts, vocal isolation, speaker diarization, and AI watermarking. GPU detection covers CUDA, MPS, ROCm, and CPU, with remote workers for distributed model downloads. The MCP server exposes synthesis and transcription tools to other applications, and diagnostics provide logs and support bundles.
VoiceStudio processes audio and video locally, using short reference clips for zero-shot cloning and text attributes for voice design. The dubbing pipeline transcribes, translates, and re-voices video while preserving speakers and timing. Stories mode imports EPUB and PDF files and exports M4B audiobooks with chapter rendering. The dictation widget provides live transcription with optional LLM cleanup, and the batch queue handles large sets of jobs or watches folders for new videos. A model catalogue manages engine installation and routing, while remote workers can receive models with live progress.
VoiceStudio includes a Rust loopback sidecar for dictation control, a local MCP server for external clients, and a watermark system based on AudioSeal. The program supports Linux ARM64 on Asahi Apple Silicon through a Vulkan-accelerated GGUF engine. Installation uses a one-command script per platform, and Docker Compose worker services allow headless GPU machines to join without a published UI. The backend reports startup progress through a dedicated endpoint, and the bug reporter detects outdated builds and offers the latest release.
| Zero-shot voice cloning from three-second reference without speaker training | Attribute-based voice synthesis from gender, age, accent, and emotion |
| Speaker-preserving video dubbing with retryable translation segment cleanup | Multi-voice EPUB and PDF rendering into chaptered M4B audiobooks |
| System-wide dictation with local LLM cleanup and clipboard restoration | Demucs-based vocal isolation extracts clean speech from mixed audio |
| Pyannote and WhisperX speaker diarization for per-speaker processing pipelines | Parallel batch queue with configurable per-engine segment width |
| Unified model catalogue routes TTS, ASR, and LLM across engines | Remote worker model installation with live startup progress tracking |
| CUDA, MPS, ROCm, and CPU device auto-detection with override | AudioSeal invisible watermarking for generated speech provenance verification |
| Versioned MCP transports expose synthesis tools to external clients | Scrubbed support bundles with outdated-build detection and crash diagnostics |
| Persistent encoded voice references skip re-encoding across restarts | Phonetic pronunciation control via pinyin or CMU phoneme input |
| Browser-safe H.264/AAC normalization prevents VP9 or AV1 preview failures | FlashInfer acceleration delivers 2.2x CUDA speedup with fallback |
| GPU exhaustion errors sanitized into actionable dubbing messages | Dependency-free CLI bridge supports Herdr, agents, and TUIs |