Quran-Lab

/

zipformer_p-arabic-v3

Like 36

Follow

Quran Lab 42

Wa jazaakAllahu Khairan, baarakAllahu feekum

ربنا تقبل منا انك انت السميع العليم
May Allah Accept <3

you can also take a look at this package ..it has a good stable offline Ayah finder , Recitation checker and tajweed checking using this model

https://github.com/Iam-Muslim/ReciteQuran

and here is an example which works on web - android - ios - windows elhamdule Allah

https://github.com/Iam-Muslim/Natlu

sa again,

For my use case your model the best so far.
my application is to recognize in a mosque the recited Ayah and give back arabic text and translation.
this is a difficult task if considered the sound path: Reciter-->Microphpne-->Loud speaker-->Phone Microphone.
however there are many cases where the sound is not recognized
I can hear the sound very well. recitation is good and understandable.
However no recognizing.
I have around 500 sound samples recorded at different mosques in Turkey.
would you like to have them to use for analysis or training of your model?
I can even record more for you.

Interested?
Best regards,
Kasım YAzan

Ahlan Akhi,

We would be very happy to receive any audio recordings you are willing to share, especially if they come with transcripts or information about the corresponding surah and ayat.

For our current work, we are particularly happy to find recordings with relatively little reverb, or a fairly dry sound, similar to the example I am sending from Shaykh Mansour Al-Salimi.

The kind of recordings we are currently trying to avoid for training are those where the reverberation is quite deeply embedded in the voice, similar to the Shaykh Mishary example I am also sending. These recordings can still be very useful for analysis and evaluation, but they are less ideal for the clean training data we are currently looking for.

That said, please feel free to send whatever you have. We would be very happy to receive the recordings and take a look at them, including the mosque recordings you mentioned.

Also, if you encounter any particular cases where the model struggles or behaves unexpectedly, please don’t hesitate to let us know and send us the sample. Those specific failure cases are very useful for us to investigate.

JazakAllahu khairan for your kind offer and willingness to help. May Allah bless you, reward you abundantly, and put barakah in your work 🤍

Best regards,
Raufa
Quran Lab

And just to clarify, Akhi, the mosque recordings you mentioned are still very valuable to us even if they contain noticeable reverb. They can be especially useful for testing, robustness evaluation, and understanding real-world failure cases.

The low-reverb / dry recordings I mentioned are mainly what we are currently looking for as cleaner training data. So please don’t hesitate to send the mosque samples as well. We would be very happy to study them, in shaa Allah

Brother thanks for the detailed answer. I do not know much about sound quality. I can not judge the reverb levels. Thats the sound in a mosque my samsung s23 can record. İt is what it is.

About the correlation of sounds to ayahs, I dont have it at the moment, even if I had it I can not garantee they are correct as I am not a hafiz.

I can make them go through Cursor/ai bu I think you have better AI tools to make the correlations.

How can I transfer the files to you?

We appreicate any sound quality barakallah feek.

You can send it to me on mostafa@quranlab.ai or raufa@quranlab.ai via Google Drive, wetransfer, dropbox, or any preferred method

jazakallah khayr

Barakallah feek. I have received the email. We will publish this and attribute it to you.

We will likely use it in our training. May Allah ﷻ multiply your sadaqah and whom who recited the words of Allah ﷻ