System Detail
wespeaker/ecapa512-voxceleb
wespeaker_ecapa512
Submitted by: Public checkpoint
- Global position
- #21 / 25
- Scenario-macro EER
- 14.3120
- Scenario-macro minDCF
- 0.541141
- Trial-set coverage
- 26 / 26
- Scenario top-three
- 0
Publicly ranked
This result is included in the reviewed public ranking.
Model Provenance
Source and architecture
- Displayed name:
wespeaker/ecapa512-voxceleb - Submitted by: Public checkpoint
- Metadata: OpenSVBench model record
- Source: WeSpeaker
- Architecture: ECAPA-TDNN
- Model family: ECAPA
- Submitted alias:
wespeaker-voxceleb-ecapa-tdnn512-LM - Created at:
2026-05-29T14:47:10.548863+00:00
Training And Links
Training data and references
- Training data: VoxCeleb2 dev, 5,994 speakers
- Training setup: ECAPA-TDNN GLOB c512 with ASTP pooling, 192-d embedding, ArcMargin, large-margin fine-tuning, speed perturbation, and MUSAN/RIRS augmentation.
- Score processing: None
- Submission mode:
full-core - Model source: Open link
- Paper: ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification
Higher-Ranked Scenarios
Scenario ranks above the global position
Show 3 supporting trial sets
- Lombard Grid Lombard: #15 on its trial-set ranking.
- GLOBE: #15 on its trial-set ranking.
- ESD: #16 on its trial-set ranking.
Lower-Ranked Scenarios
Scenario ranks below the global position
Show 3 supporting trial sets
- 3D-Speaker Distance: #23 on its trial-set ranking.
- Whisper40 Whisper: #23 on its trial-set ranking.
- CN-Celeb: #23 on its trial-set ranking.
Scenario Rankings
Full ranking across scenarios
This table shows where the model sits on each scenario, using source-balanced scenario scores and the linked trial sets as evidence.
| Scenario | Rank | Lens | Evidence | Score |
|---|---|---|---|---|
|
Aging Robustness
How stable identity representations remain across age-derived and longitudinal recording gaps.
|
#19/25
Needs work
|
Speaker aging and time gaps
3.0335 EER from leader
|
4.8252
0.212812 minDCF
|
|
|
Cross-Lingual Robustness
How well speaker identity survives enrollment-test language mismatch.
|
#19/25
Needs work
|
Language mismatch
4.8506 EER from leader
|
6.5083
0.405096 minDCF
|
|
|
Noise / Reverb Robustness
How reliably the model preserves identity when room acoustics, distractor noise, and microphone placement deviate from an easier in-corpus reference condition.
|
#19/25
Needs work
|
Noise and reverberation
5.3560 EER from leader
|
7.6160
0.264660 minDCF
|
|
|
Speaking-Style Robustness
Whether the model can preserve identity across emotion-driven change, whispered speech, and noise-induced Lombard speaking style.
|
#20/25
Needs work
|
Speaking style shift
5.1962 EER from leader
|
7.1128
0.367138 minDCF
|
|
|
Short-Duration Robustness
How much performance holds up when speech evidence is limited by duration.
|
#20/25
Needs work
|
Short-duration speech
7.9596 EER from leader
|
16.5541
0.604686 minDCF
|
|
|
Overlap Robustness
How well speaker identity survives light, mid, and heavy overlap in both meeting and domestic recordings.
|
#21/25
Needs work
|
Overlapping speakers
7.2572 EER from leader
|
14.9489
0.650355 minDCF
|
|
|
Distance Robustness
How much performance changes across meeting and domestic distance mismatch.
|
#21/25
Needs work
|
Distance mismatch
11.3505 EER from leader
|
16.0447
0.664732 minDCF
|
|
|
Accent / Dialect Robustness
How reliably the model tracks identity across accent and dialect mismatch.
|
#21/25
Needs work
|
Accent and dialect variation
9.5713 EER from leader
|
25.3328
0.762541 minDCF
|
|
|
In-The-Wild Robustness
How strong the model is on unconstrained celebrity and media speech across official CN-Celeb and VoxCeleb protocols.
|
#22/25
Needs work
|
Open-domain media speech
6.5342 EER from leader
|
8.9203
0.324306 minDCF
|
|
|
Channel / Device Robustness
How resilient the model is to device and channel mismatch.
|
#22/25
Needs work
|
Device and channel variation
14.7481 EER from leader
|
18.7142
0.846298 minDCF
|
|
|
Genre-Shift Robustness
Whether performance holds when CN-Celeb enrollment and test speech come from different source genres.
|
#22/25
Needs work
|
Source-genre variation
17.6150 EER from leader
|
30.8550
0.849922 minDCF
|
Variant Compare
Compare ECAPA variants
Compare the family variants on the overall board, a Scenario, or a Trial Set.
Show
Variant Compare
Compare ECAPA variants
Compare the family variants on the overall board, a Scenario, or a Trial Set.
| Variant | Submitted by | Source | Training Data | Training Setup | Rank | EER | minDCF | Open |
|---|---|---|---|---|---|---|---|---|
|
speechbrain/ecapa
ECAPA-TDNN
|
Public checkpoint | SpeechBrain | VoxCeleb1 + VoxCeleb2 training data | SpeechBrain ECAPA-TDNN release with attentive statistical pooling and Additive Margin Softmax loss. | #18 | 13.5461 | 0.543660 | Open |
|
wespeaker/ecapa1024-voxceleb
ECAPA-TDNN
|
Public checkpoint | WeSpeaker | VoxCeleb2 dev, 5,994 speakers | ECAPA-TDNN GLOB c1024 with ASTP pooling, 192-d embedding, ArcMargin, 150-epoch training, speed perturbation, and MUSAN/RIRS augmentation. | #19 | 14.0835 | 0.534111 | Open |
|
wespeaker/ecapa512-voxceleb
ECAPA-TDNN
Current
|
Public checkpoint | WeSpeaker | VoxCeleb2 dev, 5,994 speakers | ECAPA-TDNN GLOB c512 with ASTP pooling, 192-d embedding, ArcMargin, large-margin fine-tuning, speed perturbation, and MUSAN/RIRS augmentation. | #21 | 14.3120 | 0.541141 | Open |
Trial-Set Results
Raw ranking by trial set
Expand this section to inspect the detailed trial-set rankings.
Show
Raw ranking by trial set
Expand this section to inspect the detailed trial-set rankings.
| Trial Set | Rank | Scenarios | Trials | Score |
|---|---|---|---|---|
|
TidyVoiceX2-ASV
TidyVoiceX2-ASV
|
#19/25
|
Cross-lingual
|
200000 |
6.5083
0.405096 minDCF
|
|
HI-MIA
HI-MIA
|
#21/25
|
Short-duration
|
660000 |
8.5340
0.404212 minDCF
|
|
GSC Short
Google Speech Commands
|
#17/25
|
Short-duration
|
220000 |
12.8770
0.644885 minDCF
|
|
GLOBE
GLOBE
|
#15/25
|
Accent/dialect
|
56848 |
25.4257
0.525755 minDCF
|
|
CN-Celeb
CN-Celeb
|
#23/25
|
In-the-wild
|
3484292 |
16.3391
0.550287 minDCF
|
|
CN-Celeb Genre
CN-Celeb
|
#22/25
|
Genre shift
|
440000 |
30.8550
0.849922 minDCF
|
|
CN-Celeb Short
CN-Celeb
|
#22/25
|
Short-duration
|
546964 |
29.6374
0.854863 minDCF
|
|
VoxCeleb1-O
VoxCeleb1
|
#19/25
|
In-the-wild
|
37611 |
1.0371
0.085403 minDCF
|
|
VoxCeleb1-E
VoxCeleb1
|
#19/25
|
In-the-wild
|
579818 |
1.2104
0.077313 minDCF
|
|
VoxCeleb1-H
VoxCeleb1
|
#19/25
|
In-the-wild
|
550894 |
2.2570
0.132259 minDCF
|
|
VoxCeleb Short
VoxCeleb1
|
#18/25
|
Short-duration
|
394724 |
15.1680
0.514784 minDCF
|
|
3D-Speaker Device
3D-Speaker
|
#22/25
|
Channel/device
|
180000 |
25.3867
0.995540 minDCF
|
|
3D-Speaker Distance
3D-Speaker
|
#23/25
|
Distance
|
175163 |
25.4467
0.989464 minDCF
|
|
3D-Speaker Dialect
3D-Speaker
|
#21/25
|
Accent/dialect
|
180000 |
25.2400
0.999327 minDCF
|
|
FFSVC 2022 Cross-Channel
FFSVC 2022
|
#20/25
|
Channel/device
|
72000 |
12.0417
0.697056 minDCF
|
|
FFSVC 2022 Cross-Domain
FFSVC 2022
|
#20/25
|
Distance
|
66546 |
12.1783
0.703849 minDCF
|
|
Whisper40 Whisper
Whisper40
|
#23/25
|
Speaking style
|
17600 |
16.6250
0.780062 minDCF
|
|
Lombard Grid Lombard
Lombard Grid
|
#15/25
|
Speaking style
|
29524 |
0.2235
0.015499 minDCF
|
|
VOiCES Noise/Reverb
VOiCES
|
#19/25
|
Noise/reverb
|
55000 |
7.6160
0.264660 minDCF
|
|
CHiME-6 Domestic Far-Field
CHiME-6
|
#19/25
|
Distance
|
36487 |
15.1040
0.585680 minDCF
|
|
CHiME-6 Overlap
CHiME-6
|
#19/25
|
Overlap
|
39600 |
18.2778
0.716417 minDCF
|
|
ESD
ESD
|
#16/25
|
Speaking style
|
437408 |
4.4900
0.305853 minDCF
|
|
AliMeeting Near/Far
AliMeeting
|
#23/25
|
Distance
|
220000 |
11.4500
0.379935 minDCF
|
|
AliMeeting Overlap
AliMeeting
|
#23/25
|
Overlap
|
165000 |
11.6200
0.584293 minDCF
|
|
VoxKnesset
VoxKnesset
|
#18/25
|
Aging
|
158312 |
6.6861
0.282060 minDCF
|
|
VoxPopuli Aging
VoxPopuli
|
#19/25
|
Aging
|
146575 |
2.9644
0.143565 minDCF
|