| --- |
| title: Chatterbox TTS API |
| emoji: 🗣️ |
| colorFrom: indigo |
| colorTo: purple |
| sdk: docker |
| app_port: 7860 |
| pinned: false |
| --- |
| |
| # Chatterbox TTS API |
|
|
| CPU-only FastAPI wrapper around [ResembleAI/chatterbox](https://huggingface.co/ResembleAI/chatterbox) multilingual TTS. |
|
|
| ## Setup |
|
|
| Space → Settings → Variables and secrets → add secret `API_KEY`. |
|
|
| ## Usage |
|
|
| ```bash |
| curl -X POST https://<user>-<space>.hf.space/tts \ |
| -H "x-api-key: $API_KEY" -H "content-type: application/json" \ |
| -d '{"text": "Hello there", "lang": "en"}' |
| # {"url": "https://<user>-<space>.hf.space/audio/<uuid>.wav", "lang": "en"} |
| ``` |
|
|
| Then `curl -O <url>` — the download endpoint is public (unguessable uuid). |
|
|
| Params: `text` (required, ≤1000 chars), `lang` (default `en`), `exaggeration` (0–2), `cfg_weight` (0–1). |
|
|
| Output is 24kHz mono WAV and always carries Resemble's Perth watermark. Defaults suit most |
| prompts; for more expressive reads the model card suggests `cfg_weight` ~0.3 with |
| `exaggeration` ~0.7. Caveat: upstream `generate()` hardcodes `max_new_tokens=1000` (~40s of |
| audio), so text beyond roughly 600-700 chars is silently truncated. |
| Supported langs: ar, da, de, el, en, es, fi, fr, he, hi, it, ja, ko, ms, nl, no, pl, pt, ru, sv, sw, tr, zh. |
|
|
| ## Notes |
|
|
| - First request downloads the model (~1–2 min); CPU synthesis takes roughly 10–40s per sentence. Free Spaces sleep when idle, so the first call after a wake is slow. |
| - Audio lives on the ephemeral container disk — links die on restart. Add S3/Blob upload if you need durable URLs. |
|
|
