| --- |
| title: Audio8 TTS Preview 0.6B |
| colorFrom: red |
| colorTo: gray |
| sdk: docker |
| app_port: 7860 |
| models: |
| - Audio8/Audio8-TTS-Preview-0.6b |
| license: apache-2.0 |
| --- |
| |
| # Audio8 TTS Preview 0.6B |
|
|
| Multilingual text-to-speech and zero-shot voice cloning for |
| [`Audio8/Audio8-TTS-Preview-0.6b`](https://huggingface.co/Audio8/Audio8-TTS-Preview-0.6b). |
| The service uses the official Audio8 adapter with SGLang Omni 0.1.0 and |
| SGLang 0.5.8. |
|
|
| Reference audio uploaded through the web interface is limited to 30 seconds. |
| The A10G runtime enables `torch.compile` with SGLang's native batch policy and |
| uses CUDA Graph without an Audio8-specific compile batch-size limit. |
| Waveform decoding runs once after acoustic-token generation; incremental |
| waveform streaming is disabled for long-form stability. |
|
|
| ## API |
|
|
| ```bash |
| curl -sS \ |
| -H 'Content-Type: application/json' \ |
| -d '{ |
| "model": "audio8/tts-0.6b", |
| "input": "Audio8 generates this sentence with the reference voice.", |
| "response_format": "wav", |
| "temperature": 0.8, |
| "top_p": 0.95, |
| "top_k": 50, |
| "references": [{ |
| "audio_path": "https://example.com/reference.wav", |
| "text": "The exact transcript spoken in the reference audio." |
| }] |
| }' \ |
| https://audio8-audio8-tts-preview-0-6b.hf.space/v1/audio/speech \ |
| -o audio8-clone.wav |
| ``` |
|
|
| Use reference voices only with the speaker's permission and disclose synthetic |
| audio where appropriate. |
|
|
