Speech studio
Type a script, pick a voice and an engine, render to WAV. Compare engines side by side.
open source · local by default
A speech lab that runs on your machine.
Text to speech, transcription, cloning and voice design with 34 adapters for Hugging Face models. One desktop app for Linux, macOS and Windows. One local API for everything else.
install
$
curl -fsSL https://voice-labs.sh/install | shInstaller endpoints are placeholders until the first release is cut.
Workflows
desktopVoice clone
Drop in a short reference clip. Clean it up, then clone with any adapter that supports it.
Voice design
Describe a speaker in a sentence. Age, accent, pace, recording quality.
Transcripts
Audio or video in, editable text with timestamps out. Batch folders overnight.
Model lab
Pull weights from Hugging Face, check hardware compatibility, benchmark latency locally.
Local API
OpenAI-compatible /v1/audio endpoints on localhost. Point existing tools at it.
Hardware routing you can inspect.
CUDA, ROCm, Apple Silicon and CPU. The app checks what an adapter can run on before you pick it. The catalog shows the same information here.
developers
OpenAI-compatible local API
POST http://127.0.0.1:8471/v1/audio/speech
POST http://127.0.0.1:8471/v1/audio/transcriptions
GET http://127.0.0.1:8471/v1/models models
Hugging Face first
Weights are pulled from the Hub into a local cache you control. No account, no telemetry. Licenses are listed per adapter so you can decide what fits your use.