The Qatar Computing Research Institute (QCRI) has released QASR, a 2,000-hour transcribed Arabic speech corpus collected from Aljazeera news broadcasts. The dataset features multi-dialect speech sampled at 16kHz, aligned with lightly supervised transcriptions and linguistically motivated segmentation. QCRI also released a 130M word dataset to improve language model training. Why it matters: QASR enables new research in Arabic speech recognition, dialect identification, punctuation restoration, and other NLP tasks for spoken data.
This paper benchmarks the performance of OpenAI's Whisper model on diverse Arabic speech recognition tasks, using publicly available data and novel dialect evaluation sets. The study explores zero-shot, few-shot, and full finetuning scenarios. Results indicate that while Whisper outperforms XLS-R models in zero-shot settings on standard datasets, its performance drops significantly when applied to unseen Arabic dialects.
The Qatar Computing Research Institute (QCRI) has released SpokenNativQA, a multilingual spoken question-answering dataset for evaluating LLMs in conversational settings. The dataset contains 33,000 naturally spoken questions and answers across multiple languages, including low-resource and dialect-rich languages. It aims to address the limitations of text-based QA datasets by incorporating speech variability, accents, and linguistic diversity. Why it matters: This benchmark enables more robust evaluation of LLMs in speech-based interactions, particularly for Arabic dialects and other low-resource languages.
Researchers created Masader, the largest public catalog for Arabic NLP datasets, containing 200 datasets annotated with 25 attributes. They developed a metadata annotation strategy applicable to other languages. The paper highlights issues within current Arabic NLP datasets and suggests recommendations. Why it matters: This curated dataset directory helps lower the barrier to entry for Arabic NLP research and development.
LAraBench introduces a benchmark for Arabic NLP and speech processing, evaluating LLMs like GPT-3.5-turbo, GPT-4, BLOOMZ, Jais-13b-chat, Whisper, and USM. The benchmark covers 33 tasks across 61 datasets, using zero-shot and few-shot learning techniques. Results show that SOTA models generally outperform LLMs in zero-shot settings, though larger LLMs with few-shot learning reduce the gap. Why it matters: This benchmark helps assess and improve the performance of LLMs on Arabic language tasks, highlighting areas where specialized models still excel.