A benchmark for open-source multi-dialect Arabic ASR models
Display and explore model leaderboards and chat history
Provide feedback on text content
Playground for NuExtract-v1.5
Analyze content to detect triggers
Retrieve news articles based on a query
Generate keywords from text
Encode and decode Hindi text using BPE
fake news detection using distilbert trained on liar dataset
Generate answers by querying text in uploaded documents
Identify AI-generated text
Detect harms and risks with Granite Guardian 3.1 8B
Track, rank and evaluate open LLMs and chatbots
The Open Universal Arabic ASR Leaderboard is a benchmark platform designed to evaluate and compare open-source Arabic Automatic Speech Recognition (ASR) models. It focuses on multi-dialect Arabic speech recognition, providing a unified space for researchers and developers to assess model performance across diverse dialects and scenarios. The leaderboard ensures transparency and consistency in model evaluation, fostering innovation and collaboration in the field of Arabic speech recognition.
• Multi-dialect support: Evaluates ASR models across various Arabic dialects, including Modern Standard Arabic (MSA) and colloquial dialects. • Open-source focus: Promotes the use of open-source models to encourage community-driven improvements. • Interactive web interface: Allows users to explore benchmark results, compare models, and visualize performance metrics. • Request functionality: Users can submit requests for benchmarking new or custom Arabic ASR models. • Comprehensive metrics: Provides detailed performance metrics, such as Word Error Rate (WER), Character Error Rate (CER), and accuracy scores. • Model comparison tools: Enables side-by-side comparison of different models based on dialect, accuracy, and performance.
1. What dialects are supported on the leaderboard?
The Open Universal Arabic ASR Leaderboard supports a wide range of Arabic dialects, including Modern Standard Arabic (MSA), Egyptian, Levantine, Gulf, Iraqi, and Moroccan dialects.
2. How often are the benchmarks updated?
Benchmarks are updated periodically as new models are submitted or existing models are re-evaluated. Updates are typically announced on the platform’s news section or social media channels.
3. Can I contribute my own ASR model to the leaderboard?
Yes, the leaderboard encourages contributions from the community. You can submit your model for benchmarking by following the submission guidelines provided on the platform.