
VoiceStudio
Fully local voice cloning, dubbing, dictation, and transcription in one desktop app - an open-source ElevenLabs replacement.
VoiceStudio is a local-first voice AI workstation that packs voice cloning, voice design, video dubbing, dictation, transcription, and audiobook creation into a single desktop app. It advertises support for 646 languages and runs all speech models on your own hardware — nothing leaves your machine unless you opt in to remote providers or analytics.
The app ships as an Electron desktop client for macOS (Apple Silicon and Intel), Windows, and Linux, with official AppImage and deb packages. For headless servers there is a Docker image (ghcr.io/debpalash/voicestudio, amd64, with ROCm variants) that serves the same UI over HTTP, so the backend can run on a GPU box while you open the interface in any browser.
Under the hood it uses a Python backend with swappable engines: speech generation defaults to k2-fsa OmniVoice with alternatives like CosyVoice 3, IndexTTS 2.5, MLX-Audio, GPT-SoVITS, and sherpa-onnx, while transcription defaults to WhisperX with Faster-Whisper, MLX Whisper, Parakeet, Moonshine, and FunASR also available. GPU auto-detect covers NVIDIA CUDA, Apple Metal/MLX, and AMD ROCm. A bundled Rust sidecar handles microphone capture and text insertion for the dictation widget.
It also works as a headless speech service: a local HTTP/WebSocket API on port 3900 plus an MCP server lets coding agents and custom apps drive dictation, TTS, and transcription. Compared to ElevenLabs, VoiceStudio trades the convenience of a hosted API for privacy, no per-character billing, and full control over which models run — cloning quality varies by the open model you choose, and heavy workloads need a decent GPU or Apple Silicon machine.








