KVoiceBench - KSD-QA: leaderboard
Spoken Korean short-answer questions transferred from SD-QA, part of KVoiceBench (Kim et al., 2026). Accuracy in percent over 533 TTS-synthesized items; higher is better.
Metric: Accuracy (%; short-answer questions transferred from SD-QA; Korean spoken questions synthesized with TTS; single run at temperature 0.7). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Raon-Speech | 44.8 | |
| 2 | Fun-Audio-Chat | 32.1 | |
| 3 | HyperCLOVA X 8B Omni | 29.1 | |
| 4 | Qwen2.5-Omni | 28 | #1499 |
| 5 | MiniCPM-o 4.5 | 24.8 | |
| 6 | Step-Audio-2-mini | 23.3 | |
| 7 | Audio Flamingo 3 | 7.5 | |
| 8 | InteractiveOmni | 5.6 |
Interactive version: theaggregate.ai/benchmark?slug=kvoicebench-ksd-qa · How It Works · Data refreshed daily, snapshot 2026-11-12.