VoiceStudio: Local AI Voice Studio
/VoiceStudio
VoiceStudio is a local-first voice production tool that puts the full pipeline — cloning, synthesis, dubbing, transcription — on hardware you own. The appeal is obvious: no API keys, no per-character billing, no data leaving your machine. The cost is setup complexity and hardware you supply. For anyone already running a homelab or workstation with a GPU, it is worth the friction.
What is VoiceStudio?
VoiceStudio is a desktop application and local inference platform for voice cloning, speech synthesis, video dubbing, and transcription. It runs 16 text-to-speech engines and 11 speech-to-text engines on your own GPU or CPU, ships with a REST API compatible with OpenAI client libraries, and stores all projects, voices, and outputs locally by default. No account or subscription is required for the core workflow.
Core Capabilities for Audio Production
- ✓Zero-shot voice cloning from a 3-second reference clip, with 600+ TTS languages available across engine options. Feed the app any clean voice sample and synthesize new speech that preserves the original timbre.
- ✓End-to-end video dubbing: transcribe, translate, assign speakers with diarization, then re-synthesize each speaker while preserving their characteristics. Export back to video.
- ✓OpenAI-compatible REST API at localhost:3900/v1. Swap the base URL in any OpenAI client library to point at VoiceStudio — speech, transcription, and streaming all work the same way.
- ✓MCP server and skills.sh integration for AI coding agents. Agents can trigger speech synthesis or transcription through VoiceStudio's mounted MCP tools without touching a cloud API.
- ✓Apple Silicon native via MLX, CUDA via NVIDIA, and ROCm on Linux. The app auto-detects available compute and routes each engine to the fastest available device.
- ✓Local-first architecture: all audio, projects, and voice profiles stay on disk unless you explicitly enable a remote feature. The loopback backend on port 3900 never touches the network.
- ✓Built-in diagnostics: a self-check routine and scrubbed support bundle export help surface problems without sending any project data upstream.
Practical Applications
- •Content creators producing multilingual videos can dub directly in VoiceStudio rather than routing through cloud APIs, preserving speaker identity across languages.
- •Authors converting manuscripts to audiobooks get chapter-aware EPUB/PDF import, multi-voice script support, and M4B export from a single workflow.
- •Developers integrating speech into applications point existing OpenAI-compatible clients at the local backend, avoiding external API costs and latency.
- •Researchers and privacy-sensitive organizations needing complete control over audio processing keep every byte on their own infrastructure.
- •Accessibility-focused deployments can use the system-wide dictation widget to transcribe live audio and insert text through native OS paths.
Getting Started and Installation
Download a release package for your platform: macOS DMG (Apple Silicon), Windows MSI, Linux AppImage, or a Docker profile for CUDA/ROCm/CPU. On first launch the app bootstraps a managed Python environment and pulls the default model. On macOS you will need to right-click and approve the app on its first run. Intel Macs cannot run the local Python backend; use a remote backend or the Google Colab notebook instead.
Strengths
- ✓No API key, no subscription, no data sent to a remote server for the core workflow. Everything runs on hardware you control.
- ✓16 TTS engines and 11 ASR engines in one application, switchable in seconds via Ctrl/Cmd+E or the Model Catalogue.
- ✓OpenAI-compatible API means existing tooling — Python clients, curl scripts, agent integrations — works against VoiceStudio without modification.
- ✓Covers the full production chain: cloning, voice design, dubbing, dictation, transcription, and long-form audiobook rendering.
- ✓GPU auto-detection routes each engine to CUDA, MPS, or ROCm automatically, and falls back to CPU gracefully.
- ✓Docker and remote worker support let you distribute inference across machines on a LAN.
Understanding VoiceStudio's Scope and Limitations
- △Active beta — the README explicitly warns that main contains in-progress fixes and may change between releases. Use the latest tagged release for stable work.
- △Apple Silicon only on macOS. Intel Macs cannot run the local backend and must fall back to a remote worker or Colab notebook.
- △Linux requires glibc 2.39+, which excludes some LTS distributions. ROCm support is Linux-only and opt-in.
- △Some engines carry non-standard commercial licenses. OmniVoice weights are CC-BY-NC; IndexTTS 2.5 requires a Bilibili license above certain revenue thresholds. Users in commercial environments must audit each model they deploy.
- △Video dubbing is a complex multi-step pipeline and can require tuning per source video. The quality of the output depends heavily on the selected engine and the complexity of the source material.
VoiceStudio vs. Cloud-Based Services
Frequently Asked Questions
Yes. VoiceStudio runs entirely on your own hardware once models are downloaded. The desktop backend communicates over localhost, and all projects, voices, and outputs stay on your disk. Network features are opt-in and analytics only activate with explicit consent.
The application is licensed under AGPL-3.0, but downloaded model weights retain their upstream terms. OmniVoice uses CC-BY-NC, IndexTTS requires a separate Bilibili license for high-revenue deployments, and some engines ship under Apache-2.0 or MIT. Check each model's license before commercial use.
macOS 13.3+ on Apple Silicon, Windows 10/11 x64, and Linux x86_64 with glibc 2.39+. GPU support covers NVIDIA CUDA, Apple Silicon MPS/MLX, and ROCm on Linux. CPU-only mode works across all three platforms. Intel Macs need a remote backend.
Yes. An OpenAI-compatible REST API runs at localhost:3900/v1 for TTS and transcription. An MCP server is also available for AI coding agents like Claude Desktop or Cursor. Skills.sh integration lets agents trigger synthesis or transcription directly.
Cloud services like ElevenLabs handle infrastructure for you and get you running in minutes. VoiceStudio requires local setup and hardware, but eliminates subscription costs, per-character billing, and remote data processing. You own the entire stack once it is running.
VoiceStudio's TTS catalogue covers 646 languages, though quality varies by engine. The default OmniVoice engine supports 600+ of them. Speech-to-text works across approximately 100 languages via Whisper-based ASR engines.
The problem it solves
Commercial voice AI services lock you into per-token billing, require API keys, and send your audio to someone else's servers. For high-volume workflows, privacy-sensitive projects, or teams that want to customize every parameter, that model hits a ceiling fast. VoiceStudio fills that gap by bringing everything in-house.
Who should try it — and who should skip
VoiceStudio is worth installing if you generate a lot of audio, care about keeping data local, and have a GPU or Apple Silicon machine. The setup friction pays off in proportion to how much you use it. Skip it if you need something working in five minutes, are on an Intel Mac without a remote backend, or simply need a quick API call without managing model weights.
Related repositories
Want a second opinion on VoiceStudio?
Ask an AI that can read this page — one click and you get its take on VoiceStudio.