README.md

---
title: Chatterbox-Flash
emoji: ⚡
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 6.8.0
python_version: '3.12'
app_file: app.py
pinned: false
license: mit
short_description: Prior-calibrated block-diffusion zero-shot TTS (~2× faster)
---

# ⚡ Chatterbox-Flash — Zero-shot TTS

Gradio demo for [Chatterbox-Flash](https://github.com/resemble-ai/chatterbox-flash),
Resemble AI's prior-calibrated **block-diffusion** zero-shot TTS — the same quality as
[Chatterbox](https://github.com/resemble-ai/chatterbox) decoded in parallel (~2× faster
than a vLLM-optimized Chatterbox on an H100).

Upload a 5–10 s voice reference, type your text, and generate. Long text is split at
sentence boundaries and stitched with a short crossfade (the block-diffusion decoder can
hallucinate on very long single inputs).

## Hardware

Designed for **ZeroGPU** (the `spaces` package warm-loads the model at import and maps it
onto a GPU per request). Select **ZeroGPU** (or any CUDA GPU) in the Space's *Settings →
Hardware*. The app also runs on a standard GPU Space unchanged.

## Notes

- `sdk_version` matches what HF set when the Space was created (Gradio 6.x). If the build
  complains, bump it to a currently-available version.
- `requirements.txt` installs `chatterbox-flash` from PyPI — if it isn't published yet,
  switch that line to the `git+https://github.com/resemble-ai/chatterbox-flash.git` form.