FANAR
Arabic Speech & Language Intelligence
Speech lead for QCRI's Arabic LLM family, with a focus on dialectal Arabic, spoken language understanding, and multimodal feedback generation.
Research Scientist · Qatar Computing Research Institute
Shammur Absar Chowdhury
Conversational AI · Speech & Audio AI · Multilingual & Multimodal AI
Dr. Chowdhury specializes in designing Conversational AI models, primarily addressing complex challenges such as multispeaker interactions, nuanced multilingual and dialect variations, and code-switching, among various other intricate conversational dynamics. She is currently leading the speech technology development in Fanar — QCRI's Arabic Large Language Model project — and serves as the Lead PI on both the NAVIA and QVoice projects.
Research Interest & Expertise
Research Interests
My research focuses on conversational AI, speech and audio intelligence, and multilingual and multimodal foundation models, with an emphasis on low-resource languages, Arabic and dialectal speech, and culturally grounded evaluation.
- Conversational AI: multi-turn, multi-speaker, code-switched, and natural spoken interactions.
- Speech and audio AI: robust ASR, spoken language understanding, AudioLLMs, and expressive voice technologies.
- Multilingual and multimodal models: adaptation, benchmarking, and evaluation for underrepresented languages and cultures.
- Inclusive speech technology: tools for dialectal, accented, children’s, and neurodiverse communication.
Featured Research
Current Projects
NAVIA
AI for Neurodiverse Communication
Lead PI for an HBKU Flagship Research Grant on multimodal AI for early screening and intervention related to autism.
AURA
Audio Understanding & Representation Alignment
Audio understanding and representation alignment across acoustic, semantic, and paralinguistic signals in multilingual and low-resource settings.
QVoice
Interactive Spoken Arabic Learning
Lead PI for spoken Arabic language assessment platforms with voice-enabled input and interactive feedback for native and non-native learners.
Grants
Current Funded Research
Lead PI · 2025–2027
NAVIA: Neurodiversity Assessment and Voice-enabled Intervention AI
HBKU Flagship Research Grant, HBKU-OVPR-FRG-03-09, USD 207,000 / 750,000 QAR.
PI · 2026–2028
MediaVoice: Emotion-Aware Arabic Voice Generation for Media
QRDI-funded project on expressive Arabic voice generation for media applications.
Speech Lead
Fanar Arabic LLMs
MCIT Qatar-sponsored Arabic LLM program, with speech understanding and multimodal feedback generation.
Co-PI · 2025–2026
Native, Local and Cultural Alignment in Multilingual and Multimodal LLMs
AWS Build on Trainium support and QCRI collaboration for culturally grounded multilingual and multimodal AI.
Latest Updates
AURA Speech team is hiring Postdocs! We also have openings for interns and research assistants. If you are interested in speech and audio AI, multilingual modeling, conversational AI, or inclusive speech technology, please get in touch.
Four papers accepted at INTERSPEECH 2026:
- WASIL: In-the-Wild Arabic Spoken Interactions with LLMs
- Light-weight Pronunciation Assessment via Discrete Speech Token Surprisal
- Said Aloud, Read Different: Cross-Modal Instability in Multimodal Models
- IQRA 2026: Interspeech Challenge on Automatic Pronunciation Assessment for Modern Standard Arabic (MSA)
Co-organizing SemEval MMCultureQA 2027: a shared task on multilingual, culturally grounded spoken visual question answering at ACL 2027, Kyoto, Japan.
ACL 2026 Paper: Once Correct, Still Wrong: Counterfactual Hallucination in Multilingual Vision-Language Models (Accepted)