HADRA Moroccan Darija WebGPU ONNX

Browser-oriented ONNX conversion of Jip7e/habibi-tts-doda-darija for ONNX Runtime Web with WebGPU.

This repository does not contain a separately trained voice model. It converts the official HADRA F5-TTS weights into an encoder, FP16 transformer, and decoder used by the Amber browser client. The included F1.wav reference is Salma, the official F1 reference voice, paired with the transcript:

ู‡ูˆู…ุง ู…ุฎุจูŠูŠู† ุดูŠ ุญุงุฌุฉ ุงู†ุง ู…ุชูŠู‚ู†

Files

  • encoder_fp32.onnx
  • transformer_fp16.onnx
  • decoder_fp32.onnx
  • vocab.txt
  • F1.wav

Attribution

HADRA is an Apache-2.0 Moroccan Darija fine-tune of Habibi-TTS/F5-TTS published by the HADRA authors and trained on DODa Moroccan Arabic speech. See the upstream model card for architecture, dataset, usage, and citation details.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for Sakuragaoka/amber-hadra-webgpu

Base model

SWivid/F5-TTS
Quantized
(1)
this model