Results

Results on ASR Tasks

Models	MDRM-test	SPGISpeech-test	Earning-21	Earning-22
Whisper-v3	2.14	2.88	11.85	15.93
Qwen2-Audio-7B	3.97	4.42	26.06	42.76
Qwen2-Audio-7B-Instruct	4.68	5.74	29.58	33.65
SALMONN-7B	51.52	39.51	83.20	88.50
SALMONN-13B	49.17	41.17	80.54	86.23
Gemini-1.5-flash	4.850	5.802	18.58	27.13
Gemini-2.0-flash	4.321	5.143	19.17	28.12
GPT-4o-audio-transcribe	4.23	4.66	15.78	21.37

Results on summarization task

Metrics	Whisper-v3	Qwen2-Audio-7B-Instruct	Gemini-1.5-flash	GPT-4o-audio
Rouge-L	0.053	0.048	0.072	0.063
BertScore	0.514	0.467	0.553	0.508