NAMAA-Space/cohere-speech-tashkeel-2b

Arabic Speech Diacritization (Tashkeel)

This Space demonstrates NAMAA-Space/Cohere-Speech-Tashkeel-2B, a ~2B parameter Arabic ASR model that produces fully-diacritized text directly from speech.

Upload or record Arabic audio, and the model returns text with all ḥarakāt (fatḥa, ḍamma, kasra, sukūn, shadda, tanwīn) and grammatical case-endings (iʿrāb), inferred from the actual pronunciation.