Spaces:
Running
Licence terms for the IqraEval corpora
As-Salam alaykum wa Rahmat Allah wa Barakatuhu..
Dear IqraEval organisers,
Thank you for releasing these datasets — I'm using them to build an Arabic phoneme-level mispronunciation detection system for Qur'an recitation.
I could not find a licence file on the dataset cards for iqra_train, iqra_tts, iqra_dev or QuranMB, so I would be grateful for clarification on two points:
Under what licence are these datasets released?
What does that licence permit for a model trained on them? I would like to understand both cases, independently of what I end up doing:
Research / non-commercial use — what is permitted, and what attribution would you like?
Commercial use — is it permitted at all? If so, under what conditions (a separate agreement, a fee, attribution, or restrictions on redistributing a derived model)? If it is not permitted, that is equally useful to know.
I have no immediate plans in either direction. I'm asking now because a model trained on these corpora cannot easily be untrained later, so I would rather know both answers before that becomes a constraint.
Thank you again for making this work available.
Ahmed Khalaf
Adding a second voice to this, from a different project. We are building an offline Qur'an recitation learning device for a UK teacher, which would be sold, so the commercial question matters to us directly.
One specific point beyond Ahmed's questions: Iqra_wav2vec2_base is tagged apache-2.0 while the datasets behind it carry no licence. Is that tag intended to cover the trained weights, or the code and configuration only? If the training data turns out to be non-commercial, we would not want to rely on a permissive tag that was applied by default.
As with Ahmed, a clear "research only" would be just as useful to us as a yes.