Qwen-Audio-Jev: A Unified Interface for Audio Decisions.

arXiv:2610.21614v2 Announce Type: replace-cross Abstract: Large audio language models (LALMs) have made substantial progress in audio understanding. However, autoregressive text generation introduces computational overhead and latency in high-throughput applic