Xubo Liu

🚀 I'm hiring PhD Research Interns at Meta Superintelligence Labs (MSL) in London!

We're looking for ambitious PhD students eager to tackle challenging research problems with real-world impact! Our research sits at the intersection of audio & speech, multimodal intelligence, and LLMs.

Potential research directions include (but are not limited to):
🔹 Audio & speech representation learning, understanding and generation
🔹 Multimodal RL and post-training for audio/speech LLMs
🔹 Interaction models and real-time conversational intelligence (e.g., full-duplex)
🔹 Multimodal agents and Recursive Self-Improvement (RSI)
🔹 Evaluation and benchmarking of critical but hard-to-measure capabilities

🔗 Apply here: metacareers.com/profile/job_de…

Interested? Feel free to DM me on Linkedin or email me at xuboliu@meta.com with your CV, a brief note on your research interests, and links to relevant publications or projects!

Honored to be recognized as a Top Area Chair (AC) at NeurIPS 2026! Thanks to all the reviewers and organizers for their hard work!

Sharing our work from last summer with @aojunyi: Spoken Language Models that Think Aloud.

When humans need a moment to think, we rarely go completely silent. We might acknowledge the question, say what we’re working out, or give a quick progress update.

For reasoning-based spoken language models (SLMs), however, thinking and speaking are typically sequential: the model completes its reasoning before producing a response, which can lead to long silent gaps and less responsive voice interaction.

Our idea is simple: let SLMs think aloud asynchronously while reasoning, more like humans do. The model generates short, natural spoken responses that evolve with its reasoning, while the main reasoning process continues in the background. Same reasoning performance, more responsive voice interaction.

We’ll present this work at IEEE SLT 2026 in Palermo, Sicily this December. See you there!

📄 arxiv.org/abs/2609.26488

Heading to Interspeech 2026 in Sydney! I’ll be co-hosting Meta’s networking mixer this year.

If you’re around and want to chat about speech, audio, voice, or multimodal AI, come say hi!

events.atmeta.com/metanetworking…

Meta is hosting a networking mixer at ECCV on Sep 10, and I’ll be co-hosting!

If you’re around and want to chat about our mission, the recent Muse releases, or research in multimodal AI, audio-visual, audio, and speech, come find me. Looking forward to the conversations!

eccvnetworkingevent2026.splashthat.com