Chatmo

Direct local playground for Chatterbox Persian/Farsi synthesis.

Fine-tune repo: Thomcles/Chatterbox-TTS-Persian-Farsi
Fine-tune checkpoint: t3_fa.safetensors
Preferred checkpoint path: /data/models-bucket/generic/t3_fa.safetensors
The first run may download base multilingual weights from Hugging Face.

MCP tools for agents

/synthesize_speech: text-to-speech (Persian default with optional multilingual language code)
/clone_voice_speech: voice cloning from a reference clip + synthesis

Use View API → MCP in the app footer for generated tool schema and client config.

Generate Persian (Farsi) speech from text.

Persian text

exaggeration

Emotion intensity (0.5 = neutral default).

0 2

cfg_weight

Pacing — lower is faster (0.5 default).

0 2

speed

Playback speed factor.

0.5 2

Inference route

Use Local for CPU testing. Use ZeroGPU optimized on a HF ZeroGPU Space.

Local (CPU-safe) ZeroGPU optimized

Route: Local (CPU-safe)
Execution mode: always CPU
Target device: cpu
Weights path preference: /data/models-bucket/generic/t3_fa.safetensors

Generated audio

Sample code — copy and run separately

Code