Description
You want to voice a script without uploading it to a cloud service, and you want natural tone across several speakers. Drama Studio is a native macOS app that runs Fish Audio's Drama 3 text-to-speech model locally on Apple silicon, with hand-written Metal kernels driving the 35B mixture-of-experts backbone and no MLX or PyTorch at runtime.
It ships an HTTP API and an MCP server, so voice generation can plug into your own scripts and agents. The requirements are steep: Apple silicon, macOS 26 or later, about 49 GB of free disk for the 41.5 GB model plus a 1.5 GB codec, and 64 GB or more of unified memory recommended. The model uses the Fish Audio Community License, and cloning a real person's voice without consent is not permitted.
Local synthesis:weights are read on your machine and the script never leaves it.
Metal kernels:the backbone and codebook tower use hand-written Metal compute, the codec runs through Core ML.
HTTP and MCP:connect speech generation to other tools through the API or MCP.
Resumable downloads:pause and resume the model and codec, with SHA-256 checks.
First-run tuning:the first load spends 30 seconds to a minute tuning GPU kernels.
It ships an HTTP API and an MCP server, so voice generation can plug into your own scripts and agents. The requirements are steep: Apple silicon, macOS 26 or later, about 49 GB of free disk for the 41.5 GB model plus a 1.5 GB codec, and 64 GB or more of unified memory recommended. The model uses the Fish Audio Community License, and cloning a real person's voice without consent is not permitted.
Features
Local synthesis:weights are read on your machine and the script never leaves it.
Metal kernels:the backbone and codebook tower use hand-written Metal compute, the codec runs through Core ML.
HTTP and MCP:connect speech generation to other tools through the API or MCP.
Resumable downloads:pause and resume the model and codec, with SHA-256 checks.
First-run tuning:the first load spends 30 seconds to a minute tuning GPU kernels.
