Phonetic Pipeline

Quran Muaalem relies on quran-transcript to build a phonetic reference for each verse segment. Both the UI and API construct this reference before inference.

Reference generation path

In src/quran_muaalem/gradio_app.py:

• Aya(...).get_by_imlaey_words(...) selects the requested verse segment.
• quran_phonetizer(uthmani_ref, current_moshaf, remove_spaces=True) converts it to a phonetic script.
• The phonetic script is passed to Muaalem.__call__ alongside the audio.

This means a reference phonetic script is required for inference.

Why this matters for researchers

The reference generation stage defines the target labels for evaluation. If the Moshaf settings change, the expected phoneme sequence and sifat labels change as well. Always log:

• The exact Uthmani text segment
• The MoshafAttributes configuration
• The quran-transcript version

Next

• See Moshaf Attributes to understand how recitation settings affect labels.
• See Training → Pipeline Steps for the full data flow.