NADI 2026 TASK 1.3 CODE-SWITCHED AUTOMATIC SPEECH RECOGNITION

Organizer: harounelleuch; 3 submissions; NADI 2026 Subtask 1.3 evaluates Automatic Speech Recognition systems on highly code-switched Tunisian Arabic speech containing Arabic, French, and English. Given an input audio recording, participating systems must produce its verbatim transcription while preserving code-switched words. Systems are evaluated using corpus-level Word Error Rate (WER) as the primary metric and Character Error Rate (CER) as the secondary metric. Lower scores are better. The development and test data are provided through the NADI 2026 Code-Switched ASR dataset. Official NADI 2026 website: https://nadi.dlnlp.ai/2026/ Dataset: https://huggingface.co/datasets/fbougar