List your company

Spoken Short-Response Question Answering on Synthetic Spoken SRQA

89.6ARC-E AccuracyQwen2.5-Omni-7B
27.82443.86259.975.938May 26, 2026

Evaluation results

MethodLinks
Qwen2.5-Omni-7BEvaluation Protocol=Complete-audio baseline2026.0589.681.170.971.326.77170.810.27
DAPO controller (6 rewards)Evaluation Protocol=Streaming controller, Optimization=DAPO, Rewards=62026.0589.681.77169.225.97170.38.99
DAPO controller (5 rewards)Evaluation Protocol=Streaming controller, Optimization=DAPO, Rewards=52026.0589.181.769.666.424.97169.210.94
DAPO controller (4 rewards)Evaluation Protocol=Streaming controller, Optimization=DAPO, Rewards=42026.0588.981.768.465.124.670.368.510.87
Qwen2.5-Omni-7BEvaluation Protocol=Streaming controller2026.0587.880.868.663.522.87167.610.44
SFT controllerEvaluation Protocol=Streaming controller, Optimization=SFT2026.0586.378.168.660.521.671.766.19.82
Moshi + CoTEvaluation Protocol=Streaming baseline, Source=Shih et al. [31]2026.0577.759.856.156.916.157.856.6—
Audio Flamingo 3Evaluation Protocol=Complete-audio baseline2026.0575.559.442.232.710.166.347.44.97
Moshi QC + length-DPOEvaluation Protocol=Streaming baseline, Source=Shih et al. [31]2026.0565.44645.34614.756.946.719.31
GLM-4-Voice-9BEvaluation Protocol=Complete-audio baseline2026.0564.948.541.222.16.763.740.220.88
Moshi QC-SFTEvaluation Protocol=Streaming baseline, Source=Shih et al. [31]2026.0562.843.245.140.713.856.244.452.42
Moshi + CoT w/o streaming ASREvaluation Protocol=Streaming baseline, Source=Shih et al. [31]2026.0555.84450.146.312.259.944.8—
Audio Flamingo 3 + AF-ThinkEvaluation Protocol=Complete-audio baseline2026.0547.33531.621.734830.59.18
Moshi baselineEvaluation Protocol=Streaming baseline, Source=Shih et al. [31]2026.0530.221.522.823.88.742.823.4—