VoiceStudio.sh
Make voices. Tell stories. Keep the files.
- languages
- 646
- TTS + ASR engines
- 16 + 11
- desktop app + API
- 1 studio
Overview
Clone, design, dub, dictate, and build audiobooks in one open-source desktop studio. Local-first by default — no subscription, no usage meter; online services stay opt-in.
Capabilities
Own the Voice Pipeline
Clone, design, generate, dub, and dictate from one desktop studio, with the models and output under your control.
Move from Clip to Production
A multi-track dubbing workflow, audiobook editor, project library, voice gallery, and engine controls turn isolated model demos into repeatable work.
Run Where the Work Lives
Use MLX on Apple silicon, CUDA on NVIDIA, ROCm on Linux, Docker on a server — or the OpenAI-compatible API from another tool.
Versus the cloud
The Scoreboard
Same jobs, different terms. This is the table that made me build it.
| ElevenLabs | VoiceStudio | |
|---|---|---|
| Pricing | Subscription + usage meter | Free, open source — AGPL-3.0 |
| Voice cloning | 3s clip | 3s clip, zero-shot |
| Voice design | Gender, age | Gender, age, accent, pitch, style, dialect |
| Audiobooks | — | Full editor — EPUB/PDF in, .m4b out |
| Languages | Plan-dependent | 646 |
| Video dubbing | Cloud-only | Fully local |
| Where audio goes | Their servers | Your machine — online is opt-in |
| API keys | Account required | None for local work |
| GPU | N/A — it's a browser tab | CUDA · Apple Silicon · ROCm · CPU |
| Desktop app | — | macOS · Windows · Linux |
| TTS engines | 1 | 16 |
| ASR engines | 1 | 11 |
| MCP server | — | Claude, Cursor, any MCP client |
| Source | Closed | Fork it, extend it, ship it |
Will it run
The System Floor
A GPU is optional. The whole pipeline runs on CPU — slower, but yours.
| Minimum | Recommended | |
|---|---|---|
| OS | Windows 10 · macOS 13.3+ (Apple Silicon) · Ubuntu 24.04+ | Any modern 64-bit OS |
| RAM | 8 GB | 16 GB+ |
| VRAM | 4 GB — TTS auto-offloads to CPU | 8 GB+ (RTX 3060+) |
| Disk | 10 GB free | 20 GB+ SSD |
| GPU | Optional — CPU works | NVIDIA CUDA · Apple MPS · AMD ROCm (Linux) |
Engine room
Sixteen Voices In, Eleven Ears Out
One picker, Cmd/Ctrl+E from anywhere. Defaults are chosen; the rest auto-detect or install on first use.
| Count | Default | Standouts | |
|---|---|---|---|
| TTS | 16 | VoiceStudio — 600+ languages, clone + instruct | CosyVoice 3 · VoxCPM2 · IndexTTS 2.5 · MLX-Audio on Apple Silicon |
| ASR | 11 | WhisperX — ~100 languages, word-level timing | Parakeet TDT at ~10× realtime on CPU · FunASR with inline diarization |
Drop-in OpenAI/ElevenLabs-compatible API: point your existing SDK at localhost — voice takes your own cloned-profile IDs, model pins an engine per request.
