DIHARD III
Models identify who speaks when across eleven recording domains. The benchmark includes overlapping speech.
- Made by
- Ryant and colleagues
- Released
- September 2020
- Kind
- Audio understanding and voice
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Interaction
- Single turn
- Items from
- Real work and use
- Refresh
- New versions replace it
- Run by
- Anyone
- Based on
- A version of another benchmark
- In views
- Audio and speech
USTC-NELSLIP System tops the leaderboard with 13.4%
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. The leaderboard ranks systems by diarization error rate on the core evaluation set. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.
- 1 USTC-NELSLIP System Reference speech segmentation 13.4%
- 2 Hitachi-JHU DIHARD III System Reference speech segmentation 14.1%
- 3 Primary system W/ FE & OSD Reference speech segmentation 15.4%
- 4 DKU-Duke-Lenovo System Reference speech segmentation 15.4%
- 5 BUT DIHARD III System Reference speech segmentation 15.5%
- 6 CRIM primary system Reference speech segmentation 16.4%
- 7 WPS (FTDNN)+NES Reference speech segmentation 18.1%
- 8 EDA-EEND with conformer encoders Reference speech segmentation 18.4%
- 9 ABSP System Reference speech segmentation 18.7%
- 10 Domain Adaptive Diarization System Reference speech segmentation 19.8%
- 11 DiaboLIIc Reference speech segmentation 20.6%
- 12 DIHARD III baseline Reference speech segmentation 20.6%
- 13 AHC Reference speech segmentation 21.8%
- 14 XMUSPEECH System Reference speech segmentation 22.1%
- 15 DenseNet with agglomerative hierarchical clustering Reference speech segmentation 22.9%
- 16 GREAT-SHU Speaker Diarization System Reference speech segmentation 26.3%
- 17 Xvectors with adaptive threshold Reference speech segmentation 33.1%
- 18 UWB-NTIS x-vector clustering system Reference speech segmentation 49.4%
Source: Official DIHARD III leaderboard
| # | Entry | Score | Setting | Submitted by | Date | Link | Source |
|---|---|---|---|---|---|---|---|
| 1 | USTC-NELSLIP System | 13.4% | Reference speech segmentation | USTC-NELSLIP | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 2 | Hitachi-JHU DIHARD III System | 14.1% | Reference speech segmentation | dh_Hitachi-JHU | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 3 | Primary system W/ FE & OSD | 15.4% | Reference speech segmentation | Clova | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 4 | DKU-Duke-Lenovo System | 15.4% | Reference speech segmentation | DKU-Duke-Lenovo | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 5 | BUT DIHARD III System | 15.5% | Reference speech segmentation | dh_BUT | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 6 | CRIM primary system | 16.4% | Reference speech segmentation | CRIM | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 7 | WPS (FTDNN)+NES | 18.1% | Reference speech segmentation | dh_LEAP | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 8 | EDA-EEND with conformer encoders | 18.4% | Reference speech segmentation | dh_team_f | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 9 | ABSP System | 18.7% | Reference speech segmentation | DH_LiveFree | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 10 | Domain Adaptive Diarization System | 19.8% | Reference speech segmentation | dh_SAIL_USC | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 11 | DiaboLIIc | 20.6% | Reference speech segmentation | dh_DiaboLIIc | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 12 | DIHARD III baseline | 20.6% | Reference speech segmentation | DIHARD | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 13 | AHC | 21.8% | Reference speech segmentation | dh_ciaic | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 14 | XMUSPEECH System | 22.1% | Reference speech segmentation | XMUSPEECH | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 15 | DenseNet with agglomerative hierarchical clustering | 22.9% | Reference speech segmentation | HK | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 16 | GREAT-SHU Speaker Diarization System | 26.3% | Reference speech segmentation | dh_Great | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 17 | Xvectors with adaptive threshold | 33.1% | Reference speech segmentation | dh_sh | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
| 18 | UWB-NTIS x-vector clustering system | 49.4% | Reference speech segmentation | dh_UWB-NTIS | Dec 2020 | dihardchallenge.github.io | Official leaderboard |
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Audio and video
- Video-MMMU In use 300 expert-level lecture videos with questions testing perception, comprehension and adaptation of knowledge. Accuracy.
Qwen 3.8 Max 88.7%
- LVBench In use Hours-long video understanding with multiple-choice questions on reasoning, grounding and summarization.
Gemini 4 Argon 91.7%
- MMVU In use 3,000 expert-annotated questions on videos across 27 disciplines. Multiple-choice and open-ended accuracy.
Qwen 3.8 Max 82.4%
- MLVU In use Multi-task long video understanding (minutes to hours) with multiple-choice and generation tasks. Accuracy.
Qwen 3.8 Max 90.8%
- FLEURS In use Models transcribe read speech in 102 languages built from FLoRes-101 sentences. Model cards usually report word error rate, so lower scores are better.
Phi-4-multimodal 4.00%
- VoiceBench In use Spoken instructions with speaker, environment and content variation. Voice assistant responses scored by GPT judge or rules.
Qwen3.5-Omni-Plus 93.1%
Data audit
All 8 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
- Pass Independent runs Scores from the official leaderboard (18 entries).
- Pass Models scored 18 entries on its leaderboard, none a model we track.
- Note Newest score No scores yet.
- Note Frontier models No models scored yet.
- Note Headroom Lower is better, so there’s no ceiling to reach.
Leaderboard import 3 pass
- Pass Importer We read the board on dihardchallenge.github.io: 18 entries at the last read, none a model we track.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.