Word-by-word recitation tracking on this device uses an open Arabic speech model. Maqam does not imply endorsement by NVIDIA.
1. Acoustic model
Credit: NVIDIA
Model: FastConformer Arabic speech recognition (stt_ar_fastconformer_hybrid_large_pcd_v1.0)
License: CC BY 4.0
NVIDIA model page: huggingface.co/nvidia/stt_ar_fastconformer_hybrid_large_pcd_v1.0
2. How Maqam uses it
Maqam did not retrain this model. The weights are bundled as ONNX so tracking can run on the phone. Mixed integer quantization is from the upstream export, not a change made in Maqam.