NADI2025 Subtask2 ASR Test: Multidialectal Arabic Speech Recognition

A test set for the NADI-2015 Subtask-2 challenge focused on Automatic Speech Recognition for Arabic across multiple dialects. The dataset is hosted by UBC-NLP and was last updated in July 2025. Participants are required to register for the shared task to access the data.

Use Cases

• Evaluating zero-shot Arabic ASR model performance on dialectal speech based on the described test set
• Benchmarking fine-tuned ASR systems on multidialectal Arabic transcription tasks as part of the shared task
• Developing models to handle phonetic variation across Arabic dialects using the described training and validation data

Strengths

• Designed for a specific, peer-reviewed shared task (NADI-2015 Subtask-2) hosted on Codabench
• Focuses on a challenging real-world problem: transcribing Arabic speech across multiple dialects

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• Access requires registration for the shared task, which may be a barrier

Provenance

Source

UBC-NLP

Collection Method

Likely collected or curated for the NADI-2015 shared task subtask on multidialectal Arabic ASR.

Freshness

Last updated 2025-07-23 08:21:14

Geography

Likely covers multiple Arabic-speaking regions, given the focus on dialectal variation.