Training-Free Speech-Centric Omni Understanding with Frozen VLMs

Benchmark Model Rank Results
audio-visual-question-answering-on-futureomniQwen3-VL-30B-A3B-TFO#2Accuracy: 54.0
video-question-answering-on-videoholmesMiniCPM4.5-V-TFO-9B#5Accuracy: 59.3
video-question-answering-on-lvbenchMiniCPM4.5-V-TFO-9B#36Accuracy: 50.5