Research Outputs

Publications & Projects

Representative open-source systems, datasets, and invited talks. The complete and most current publication record is maintained on Google Scholar.

Complete Publication List

More than 60 papers in leading speech conferences and journals. Visit Google Scholar for the latest papers and citation information.

Google Scholar

Toolkits and benchmarks spanning speaker representation, target speaker extraction, music generation, and speech language model evaluation.

A comprehensive speaker embedding learning toolkit supporting industrial-scale training, research, and production deployment.

The first open-source target speaker extraction toolkit, designed for scalable and generalizable personalized speech extraction.

Fast diffusion-based rhythmic music generation with an emphasis on efficient, high-quality synthesis.

Autoregressive diffusion-based music generation for high-quality, high-fidelity, and diverse song synthesis.

A real-world, conversation-centric benchmark for target speaker extraction in natural multi-speaker interactions.

A multi-tier, multi-speaker, multilingual, multi-scenario, and multi-task benchmark for large speech language models.

2025.10 · Deep Speaker Representation Learning, APSIPA 2025 Tutorial · Slides
2025.10 · Deep Speaker Representation Learning, NCMMSC 2025 Tutorial · Slides
2025.08 · One Embedding Doesn't Fit All: Rethinking Speaker Modeling for Various Speech Applications, Interspeech 2025 MLC Workshop · Slides
2025.08 · The Real-T Dataset, Interspeech 2025 · Slides
2024.09 · Speaker Representation Learning: Theories, Applications and Practice, Brno University of Technology · Video