Spaces:
Running on Zero
Running on Zero
| title: Chatterbox-Flash TTS | |
| emoji: 🎙️ | |
| colorFrom: red | |
| colorTo: indigo | |
| sdk: gradio | |
| sdk_version: 6.26.0 | |
| app_file: app.py | |
| short_description: Block-diffusion zero-shot text-to-speech by Resemble AI | |
| python_version: "3.12" | |
| startup_duration_timeout: 30m | |
| # Chatterbox-Flash TTS | |
| [Chatterbox-Flash](https://github.com/resemble-ai/chatterbox-flash) is a prior-calibrated block-diffusion zero-shot text-to-speech model by Resemble AI. It extends Chatterbox-TTS with a parallel masked decoder that enables streaming generation with significantly lower latency. | |
| ## How it works | |
| 1. Enter the text you want to synthesize. | |
| 2. Optionally upload a short reference audio clip to clone a specific voice. | |
| 3. Click **Generate** to produce speech. | |
| The model uses a block-diffusion decoder (based on Llama-520M) with prior-calibrated PMI scoring and early decoding via a time-shifted quantile schedule, paired with the original S3Gen flow-matching vocoder and GE2E voice encoder. | |
| ## Reference | |
| - **Paper**: [Chatterbox-Flash: Prior-Calibrated Block Diffusion for Streaming Zero-Shot TTS](https://huggingface.co/papers/2605.30748) | |
| - **Model**: [ResembleAI/chatterbox-flash](https://huggingface.co/ResembleAI/chatterbox-flash) | |
| - **Code**: [github.com/resemble-ai/chatterbox-flash](https://github.com/resemble-ai/chatterbox-flash) | |
| - **License**: MIT |