Lead our work on novel audio LLMs and full-duplex audio modeling. You will design and train the next generation of our speech-to-speech models, moving beyond cascaded systems to end-to-end architectures.
The role
Responsibilities
•
Research and train novel architectures for real-time speech-to-speech translation.
•
Explore full-duplex audio models that handle interruptions, backchanneling, and overlapping speech naturally.
•
Work on voice cloning and style transfer to preserve speaker identity across languages.
•
Read papers and implement state-of-the-art techniques quickly.
Requirements
•
Deep understanding of Transformer architectures, diffusion models, and neural audio generation.
•
Experience training large-scale models from scratch.
•
A product-first research mindset — you care about end-user experience, not just benchmarks.