SEA-SpeechBench - ASR: leaderboard
Speech recognition in SEA-SpeechBench (Liao et al., 2026) across 11 Southeast Asian languages, on short clips from curated public corpora with an English instruction prompt. Raw word error rate, or character error rate for languages without word boundaries, averaged per language and then over the languages; values above 1 are possible and lower is better.
Metric: Word or character error rate (raw ratio, lower is better; WER, or CER for languages without explicit word boundaries, averaged over the source test sets of each language and then over the 11 languages (English, Filipino, Vietnamese, Indonesian, Tamil, Thai, Chinese, Khmer, Lao, Malay and Burmese); short clips of at most 30 s from curated public Southeast Asian speech corpora, up to 1,000 sampled items per dataset; English instruction prompt; exceeds 1 when a hypothesis is much longer than its reference). Source: arxiv.org. Saturation forecast: Estimated already saturated. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 0.15 | #452 |
| 2 | MERaLiON-2-10B | 0.3 | #1646 |
| 3 | MERaLiON-2-3B | 0.48 | #6258 |
| 4 | Qwen3 Omni 30B A3B Instruct | 0.56 | #902 |
| 5 | Gemma 3n E4B (IT) | 1.14 | #2220 |
| 6 | Qwen2.5-Omni-7B | 1.33 | #1797 |
| 7 | Voxtral-Mini-3B-2507 | 1.36 | #3702 |
| 8 | Gemma 3n E2B (IT) | 1.88 | #3758 |
| 9 | Phi-4 Multimodal Instruct | 2.43 | #3914 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=sea-speechbench-asr · How It Works · Data refreshed daily, snapshot 2026-11-12.