oaattia/habibi-tts-endpoint

Habibi-TTS Inference Endpoint Handler

Custom inference endpoint handler for Habibi-TTS — a zero-shot Arabic TTS model supporting 12 dialects.

Usage

Deploy this repo as a Hugging Face Inference Endpoint (Dedicated) with a GPU instance (T4 or better).

Request Format

{
  "ref_audio_url": "https://example.com/reference.wav",
  "ref_text": "Transcription of the reference audio",
  "gen_text": "النص العربي المراد تحويله إلى كلام",
  "dialect": "SAU",
  "speed": 1.0,
  "nfe_step": 32,
  "cross_fade_duration": 0.15,
  "seed": -1,
  "remove_silence": false
}

Supported Dialects

Response

Returns raw WAV audio bytes (audio/wav).