An ONNX export of tarteel-ai/whisper-base-ar-quran (Apache-2.0) for on-device Quran recitation recognition in the browser with transformers.js, used by the free تسميع (Tasme3) memorization app.
onnx/encoder_model_quantized.onnx (23 MB) and onnx/decoder_model_merged_quantized.onnx (80 MB), dynamic 8-bit weights.input_features cut to the clip length (≥ 1000 frames).No changes to the weights beyond quantization; all credit for the model to Tarteel AI.
onnx/decoder_model_merged_quantized.onnx 80 MBonnx/encoder_model_quantized.onnx 23 MBtarteel-ai/whisper-base-ar-quran