List your company

Empathetic Response Generation on AvaMERG (test)

0.2851BLEU-1ES4R
0.1122520.1571260.2020.246874Jan 16, 2026Feb 9, 2026Mar 5, 2026Mar 30, 2026Apr 23, 2026May 17, 2026Jun 11, 2026

Evaluation results

MethodLinks
ES4RBackbone=Qwen, Training=Fine-tuned2026.010.28510.16420.10950.07580.90580.33170.12220.25660.02250.1279—
ES4RBackbone=Llama, Training=Fine-tuned2026.010.27730.16030.10990.07730.90550.32590.11920.25110.02440.149—
Qwen2.5-OmniTraining=Fine-tuned2026.010.27370.15750.09520.06830.88440.30650.11290.2460.02560.1394—
BLSP-emoTraining=Fine-tuned2026.010.27080.14170.08210.05140.86880.26940.0770.21530.02190.1145—
OpenS2STraining=Fine-tuned2026.010.2690.14030.08620.05560.88340.26760.08690.21610.02110.0938—
ES4Rablation=removing speech-guided cross-modal fusion module, Training=Fine-tuned2026.010.26590.12430.07160.04480.8830.23520.06060.18320.01990.0892—
PRISMModel=PRISM (Llama), Backbone=Llama2026.060.23180.12230.08050.0555—0.20270.06490.17430.04090.25740.8801
ES4Rablation=removing the dual-level attention module, Training=Fine-tuned2026.010.22640.05520.01840.00780.87980.21670.01140.160.01690.1007—
LLaMA-Omni 2Training=Fine-tuned2026.010.22020.09110.05040.02840.87830.22460.04560.17340.03170.1571—
SALMONNTraining=Fine-tuned2026.010.20420.10520.03680.02540.85090.20980.09210.14260.02020.0908—
PRISMModel=PRISM (Qwen), Backbone=Qwen2026.060.20410.11420.07920.0571—0.22540.07450.18720.0390.25190.8792
OpenS2SModel=OpenS2S2026.060.18830.070.03550.0192—0.17590.03560.14080.04280.23290.8691
Qwen2.5-Omni-7BModel=Qwen2.5-Omni-7B2026.060.17370.08310.0530.0352—0.1880.05420.15550.03750.2380.8746
Whisper+LLMTraining=Fine-tuned2026.010.17290.06160.03260.01930.87240.18220.02590.13790.01850.0735—
PRISM (w/o Prosody-Desc)Model=w/o Prosody-Desc, Ablation=Removing prosody descriptions, Backbone=Qwen2.5-7B-Instruct2026.060.17270.05830.02710.0137—0.15210.02020.12620.0280.16170.8641
ASR+LLMTraining=Fine-tuned2026.010.17010.08010.04980.03240.86810.18260.08130.15190.0320.1978—
PRISM (Always Kno)Model=Always Kno, Ablation=Enforcing knowledge usage at every turn, Backbone=Qwen2.5-7B-Instruct2026.060.16060.0560.02590.0133—0.16110.02040.13010.02580.1550.8667
PRISM (w/o Kno)Model=w/o Kno, Ablation=Excluding knowledge entirely, Backbone=Qwen2.5-7B-Instruct2026.060.16010.05490.02540.0129—0.16240.02050.130.0250.14950.8633
LLaMA-Omni2Model=LLaMA-Omni22026.060.15650.06460.03540.0205—0.17030.03290.1330.0460.23760.8674
SALMONN-13BModel=SALMONN-13B2026.060.14640.0570.03030.0174—0.16660.03810.12890.02180.13270.8705
ASR+LLMModel=ASR+LLM2026.060.14310.05140.0250.0132—0.1690.02710.14060.02860.17220.8652
SALMONN-7BModel=SALMONN-7B2026.060.14150.06160.03570.0217—0.15980.03210.12260.02250.13460.8684
OSUM-EChatModel=OSUM-EChat2026.060.13810.04850.02260.0111—0.15460.02630.11460.03420.21330.8673
SpeechGPTModel=SpeechGPT2026.060.11890.05430.03040.018—0.14370.02280.11260.03060.1340.8534