End-to-end audio foundation model, designed for low latency and real-time conversations