System Detail
speechbrain/ecapa
ecapa
Submitted by: Public checkpoint
- Global position
- #15 / 23
- Scenario-macro EER
- 13.5461
- Scenario-macro minDCF
- 0.543660
- 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:
speechbrain/ecapa - Submitted by: Public checkpoint
- Metadata: OpenSVBench model record
- Source: SpeechBrain
- Architecture: ECAPA-TDNN
- Model family: ECAPA
- Submitted alias:
spkrec-ecapa-voxceleb - Created at:
2026-05-29T14:40:34.513468+00:00
Training And Links
Training data and references
- Training data: VoxCeleb1 + VoxCeleb2 training data
- Training setup: SpeechBrain ECAPA-TDNN release with attentive statistical pooling and Additive Margin Softmax loss.
- Submission mode:
full-core - Model source: Open link
- Paper: ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification
Training/evaluation overlap
This checkpoint was trained or fine-tuned on VoxCeleb1. Results on VoxCeleb1-E and VoxCeleb1-H are not independent of the training data.
Higher-Ranked Scenarios
Scenario ranks above the global position
Show 3 supporting trial sets
- VoxCeleb1-E: #5 on its trial-set ranking.
- VoxCeleb1-H: #7 on its trial-set ranking.
- VOiCES Noise/Reverb: #11 on its trial-set ranking.
Lower-Ranked Scenarios
Scenario ranks below the global position
Show 3 supporting trial sets
- GSC Short: #19 on its trial-set ranking.
- Whisper40 Whisper: #19 on its trial-set ranking.
- FFSVC 2022 Cross-Domain: #19 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 |
|---|---|---|---|---|
|
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.
|
#11/23
Competitive
|
Noise and reverberation
3.3200 EER from leader
|
5.5800
0.213320 minDCF
|
|
|
Accent / Dialect Robustness
How reliably the model tracks identity across accent and dialect mismatch.
|
#14/23
Needs work
|
Accent and dialect variation
6.6892 EER from leader
|
23.5109
0.730128 minDCF
|
|
|
Aging Robustness
How stable identity representations remain across age-derived and longitudinal recording gaps.
|
#15/23
Needs work
|
Speaker aging and time gaps
2.7610 EER from leader
|
4.5528
0.218235 minDCF
|
|
|
Distance Robustness
How much performance changes across meeting and domestic distance mismatch.
|
#15/23
Needs work
|
Distance mismatch
9.0732 EER from leader
|
14.3300
0.630306 minDCF
|
|
|
In-The-Wild Robustness
How strong the model is on unconstrained celebrity and media speech across official CN-Celeb and VoxCeleb protocols.
|
#16/23
Needs work
|
Open-domain media speech
4.2760 EER from leader
|
7.9977
0.317381 minDCF
|
|
|
Overlap Robustness
How well speaker identity survives light, mid, and heavy overlap in both meeting and domestic recordings.
|
#16/23
Needs work
|
Overlapping speakers
6.7429 EER from leader
|
14.8522
0.616222 minDCF
|
|
|
Channel / Device Robustness
How resilient the model is to device and channel mismatch.
|
#16/23
Needs work
|
Device and channel variation
12.6827 EER from leader
|
16.6850
0.817560 minDCF
|
|
|
Genre-Shift Robustness
Whether performance holds when CN-Celeb enrollment and test speech come from different source genres.
|
#17/23
Needs work
|
Source-genre variation
15.2965 EER from leader
|
29.5690
0.883898 minDCF
|
|
|
Speaking-Style Robustness
Whether the model can preserve identity across emotion-driven change, whispered speech, and noise-induced Lombard speaking style.
|
#18/23
Needs work
|
Speaking style shift
5.3095 EER from leader
|
7.2261
0.422165 minDCF
|
|
|
Short-Duration Robustness
How much performance holds up when speech evidence is limited by duration.
|
#18/23
Needs work
|
Short-duration speech
8.0417 EER from leader
|
17.0764
0.621298 minDCF
|
|
|
Cross-Lingual Robustness
How well speaker identity survives enrollment-test language mismatch.
|
#19/23
Needs work
|
Language mismatch
5.9699 EER from leader
|
7.6275
0.509746 minDCF
|
Variant Compare
Compare ECAPA variants
Expand this section to compare models in the same model family by architecture, source, training data, training setup, and leaderboard result.
Show
Compare ECAPA variants
Expand this section to compare models in the same model family by architecture, source, training data, training setup, and leaderboard result.
| Variant | Submitted by | Source | Training Data | Training Setup | Global Rank | Macro EER | Macro minDCF | Open |
|---|---|---|---|---|---|---|---|---|
|
speechbrain/ecapa
ECAPA-TDNN
Current
|
Public checkpoint | SpeechBrain | VoxCeleb1 + VoxCeleb2 training data | SpeechBrain ECAPA-TDNN release with attentive statistical pooling and Additive Margin Softmax loss. |
#15
ranked
|
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. |
#16
ranked
|
14.0835 | 0.534111 | Open |
|
wespeaker/ecapa512-voxceleb
ECAPA-TDNN
|
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. |
#18
ranked
|
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
|
#19/23
|
Cross-lingual
|
200000 |
7.6275
0.509746 minDCF
|
|
HI-MIA
HI-MIA
|
#15/23
|
Short-duration
|
660000 |
7.6850
0.369167 minDCF
|
|
GSC Short
Google Speech Commands
|
#19/23
|
Short-duration
|
220000 |
17.0600
0.777175 minDCF
|
|
GLOBE
GLOBE
|
#13/23
|
Accent/dialect
|
56848 |
25.4450
0.533282 minDCF
|
|
CN-Celeb
CN-Celeb
|
#17/23
|
In-the-wild
|
3484292 |
15.2177
0.579855 minDCF
|
|
CN-Celeb Genre
CN-Celeb
|
#17/23
|
Genre shift
|
440000 |
29.5690
0.883898 minDCF
|
|
CN-Celeb Short
CN-Celeb
|
#18/23
|
Short-duration
|
546964 |
28.7238
0.867746 minDCF
|
|
VoxCeleb1-O
VoxCeleb1
|
#15/23
|
In-the-wild
|
37611 |
0.9038
0.071152 minDCF
|
|
VoxCeleb1-E
VoxCeleb1
|
#5/23
|
In-the-wild
|
579818 |
0.4563
0.030992 minDCF
|
|
VoxCeleb1-H
VoxCeleb1
|
#7/23
|
In-the-wild
|
550894 |
0.9731
0.062575 minDCF
|
|
VoxCeleb Short
VoxCeleb1
|
#13/23
|
Short-duration
|
394724 |
14.8367
0.471104 minDCF
|
|
3D-Speaker Device
3D-Speaker
|
#16/23
|
Channel/device
|
180000 |
21.0200
0.926620 minDCF
|
|
3D-Speaker Distance
3D-Speaker
|
#15/23
|
Distance
|
175163 |
21.5753
0.911160 minDCF
|
|
3D-Speaker Dialect
3D-Speaker
|
#15/23
|
Accent/dialect
|
180000 |
21.5767
0.926973 minDCF
|
|
FFSVC 2022 Cross-Channel
FFSVC 2022
|
#19/23
|
Channel/device
|
72000 |
12.3500
0.708500 minDCF
|
|
FFSVC 2022 Cross-Domain
FFSVC 2022
|
#19/23
|
Distance
|
66546 |
12.4662
0.716863 minDCF
|
|
Whisper40 Whisper
Whisper40
|
#19/23
|
Speaking style
|
17600 |
15.6250
0.849062 minDCF
|
|
Lombard Grid Lombard
Lombard Grid
|
#14/23
|
Speaking style
|
29524 |
0.2608
0.021386 minDCF
|
|
VOiCES Noise/Reverb
VOiCES
|
#11/23
|
Noise/reverb
|
55000 |
5.5800
0.213320 minDCF
|
|
CHiME-6 Domestic Far-Field
CHiME-6
|
#12/23
|
Distance
|
36487 |
12.4299
0.530841 minDCF
|
|
CHiME-6 Overlap
CHiME-6
|
#15/23
|
Overlap
|
39600 |
18.1111
0.681778 minDCF
|
|
ESD
ESD
|
#16/23
|
Speaking style
|
437408 |
5.7925
0.396047 minDCF
|
|
AliMeeting Near/Far
AliMeeting
|
#17/23
|
Distance
|
220000 |
10.8485
0.362360 minDCF
|
|
AliMeeting Overlap
AliMeeting
|
#19/23
|
Overlap
|
165000 |
11.5933
0.550667 minDCF
|
|
VoxKnesset
VoxKnesset
|
#16/23
|
Aging
|
158312 |
6.9067
0.325019 minDCF
|
|
VoxPopuli Aging
VoxPopuli
|
#13/23
|
Aging
|
146575 |
2.1989
0.111452 minDCF
|