Install Gloam Voice Studio, clone your first voice, and drive it programmatically through the local API. Everything runs on your Mac.
Gloam Voice Studio is a native macOS app for Apple Silicon. You'll need:
Download the latest signed build from GitHub Releases, unzip, and drag Gloam Voice Studio.app into /Applications. On first launch, pick a synthesis model and it downloads in-app with a storage preflight.
Prefer to build it yourself? You'll need Xcode 16+ and xcodegen.
The underlying EngineKit and StudioKit packages ship unit tests — run swift test.
Cloning a voice takes one reference clip:
1. Add a reference. Record directly in the app or drop an audio file. Reference clips auto-transcribe on-device — an optional transcript hint improves quality.
2. Name & save. The voice is saved to your library as a folder containing meta.json and ref.wav.
3. Generate. Type a line in the Studio editor, choose a backend, and synthesize. Generate two A/B takes to compare waveforms side-by-side.
Every take can be directed with one of five emotion variants and a playback speed from 0.5× to 2.0×.
For capable backends, per-session direction overrides let you drive temperature (Fish S2-Pro) or exaggeration (Chatterbox) directly, overriding the emotion preset.
Voice Lab is a multi-line script editor. Each line carries its own voice and emotion direction, and lines can be reordered at will.
Generate every line in a session serially with one click — per-line status runs queued → generating → done. Export the whole script as one WAV: the starred (or newest) take per line, a configurable silence gap between lines, and optional peak normalization to −0.18 dBFS.
Swap synthesis backends per take. As a rough guide:
The Qwen3-TTS family also supports inventing a voice from a natural-language description (qwen3-design) and directing a preset speaker with instructions (qwen3-custom).
Enable the optional HTTP server in Settings to drive synthesis programmatically. It binds to localhost only, on port 8790 by default, and speaks an OpenAI-compatible surface.
Because it's OpenAI-compatible, the server can both drive and be driven by the Python sibling project gloam-voice-engine.
Voices are portable. A .gvoice pack is a ZIP archive containing meta.json and ref.wav — the shared interchange format between the macOS app and the Python engine.
Export a voice from the studio and import it into gloam-voice-engine, or vice versa. Perfect for shipping your Gloam.fm DJ voices between environments.
Everything lives inside the macOS sandbox container. Nothing is uploaded or synced.