Fun Research

Welcome to Fun Research! This repository is maintained by the Qwen Audio Team at Alibaba Group, serving as an open-source platform for our cutting-edge research in speech, audio, NLP technologies. We believe in accelerating scientific progress through transparent collaboration, and invite the global research community to explore, reproduce, and build upon our work.

Why Fun Research?

• 🚀 State-of-the-Art Models: Official implementations of our latest research breakthroughs
• 📊 Reproducibility: Pre-trained models and benchmark datasets with evaluation scripts
• 🌍 Community-Driven: Built for and with the global speech research community

Featured Projects

Explore our latest research implementations:

1. Say More with Less: Variable-Frame-Rate Speech Tokenization via Adaptive Clustering and Implicit Duration Coding, AAAI 2026 (oral)
2. ECoM-Reasoning: Efficient Chain-of-Modality Reasoning for Spoken Language Models, ACM MM 2026
3. Spoken Function Calling: A New Perspective on Spoken Language Understanding for Large Audio Language Models, ACM MM 2026
4. Agentic-GER: Terminology Recovery in Long-Form Speech Using Global Context

Disclaimer

This repository contains research artifacts:

⚠️ Not an official Alibaba product
⚠️ Released for academic/research purposes only
⚠️ Models may be subject to specific license terms