Khatma: Quran recitation model for the browser
This repository hosts the speech recognition model used by Khatma to follow Quran recitation directly on the user's device. Audio never leaves the device: the model runs in the browser through onnxruntime-web, inside a Web Worker.
Attribution
This is an adapted version of NVIDIA FastConformer-Hybrid PCD Large (ar)
(stt_ar_fastconformer_hybrid_large_pcd_v1.0), by NVIDIA:
https://huggingface.co/nvidia/stt_ar_fastconformer_hybrid_large_pcd_v1.0
Licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0): https://creativecommons.org/licenses/by/4.0/
Changes made
The original model was modified. Only the CTC head is used. The model was exported to ONNX with audio preprocessing inside the graph (the starting point was the ONNX export from the Tilawa project v0.2.0, github.com/yazinsai/tilawa, which in turn comes from github.com/Cyberistic/offline-quran-validation). It was then quantized and repacked for faster inference in the browser: int4 matrix multiplications and 1x1 convolutions were rewritten as int8 MatMulInteger, because onnxruntime-web runs the int4 path about four times slower. The package format is specific to Khatma and is unpacked in the browser before the ONNX session is created.
File
| File | Size | SHA-256 |
|---|---|---|
fastconformer_ar_khatma_v1.bin |
89 410 608 bytes | 67043116646830400106be8177f69823e70a90d9b3a5d96d6bb374cab9dc9b9f |
Requires onnxruntime-web 1.24.2 or newer (1.21 and 1.22 reject the ConvInteger operator).
Third party notices
The ONNX export used as the starting point (fastconformer_full_mixed.onnx, Tilawa release
v0.2.0) is distributed by the Tilawa project under the MIT License, together with NVIDIA's
CC BY 4.0 for the model itself ("FastConformer assets are MIT / NVIDIA CC-BY-4.0"). Tilawa's
other engine (Zipformer, NPL-1.2) is not used here. The export in turn comes from
github.com/Cyberistic/offline-quran-validation.
MIT License
Copyright (c) 2026 yazinsai
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
No warranty
The model is provided as is, without warranties of any kind, as stated in the license. Speech recognition makes mistakes, and it does not check tajweed or harakat.
Model za praćenje učenja Kur'ana u aplikaciji Khatma. Radi na telefonu, zvuk ne izlazi sa uređaja. Izvorni model je NVIDIA FastConformer (ar), licenca CC BY 4.0, izmijenjen kako je gore opisano.