Improving Real-time Quranic ASR (Whisper Small AR)
I’ve been testing the Mahmoud-Nasser/whisper-small-ar model on Quranic recitation. It works well, but I’m facing a few issues:
When the user recites, it often transcribes only 3–4 words instead of the full ayah.
The transcription response is a bit slow — I’d like it to be more immediate for real-time correction.
For longer recitations (full ayah), it doesn’t always return the complete text.
I want to improve this so the model gives faster and full transcription during Quran recitation. I’d really appreciate your guidance, and if possible, a short meeting would be very helpful.
I totally agree with you. This is the first model I have worked on, but there are other models we can test if you would like, just not on my account. Please send me your phone number so we can connect. I hope to hear your point of view.
Sorry for late response this is my contact number +92 333 8607062