System Detail
w2vbert2.0+ MHFA
w2vbert2.0+ MHFA+uncertainty
Submitted by: abc
- Global position
- #1 / 26
- Scenario-macro EER
- 5.3492
- Scenario-macro minDCF
- 0.267001
- Trial-set coverage
- 26 / 26
- Scenario top-three
- 9
Publicly ranked
This result is included in the reviewed public ranking.
Model Provenance
Source and architecture
- Displayed name:
w2vbert2.0+ MHFA - Submitted by: abc
- Metadata: Maintainer override
- Source: WeSpeaker
- Model family: W2V-BERT 2.0
- Submitted alias:
w2vbert2.0+ MHFA+uncertainty - Created at:
2026-11-15T17:57:11.785006+00:00
Training And Links
Training data and references
- Training data: DataSet Sampling Share (SpkCount, Hours, NaturalShare) VoxBlink2 25.74% (111,284 spk, 16,122 h, 45.03%) Common Voice 13.07% (67,317 spk, 5,798 h, 22.87%) CN-Celeb 9.86% (2,793 spk, 1,329 h, 2.88%) VoxCeleb2 8.52% (5,994 spk, 2,506 h, 4.96%) TidyVoiceX 8.18% (3,666 spk, 356 h, 1.19%) KeSpeech 7.53% (27,158 spk, 1,076 h, 4.39%) PrivateData 6.03% (19,236 spk, 1,260 h, 2.11%), VOiCES 5.35% (387 spk, 2,947 h, 3.12%), 3D-Speaker 4.51% (10,000 spk, 1,107 h, 2.63%), 3D-Speaker-3A 4.51% (10,000 spk, 3,339 h, 7.90%).
- Training setup: toolkit: wespeaker projection: ArcMarginProduct_intertopk_subcenter_uncertianty DataAug: noise (musan+DNS5_noise) + reverberation(DNS5_RIR+RIRS)
- Score processing: None
- Submission mode:
full-core
Higher-Ranked Scenarios
Scenario ranks above the global position
Show 3 supporting trial sets
- Lombard Grid Lombard: #1 on its trial-set ranking.
- VoxPopuli Aging: #1 on its trial-set ranking.
- TidyVoiceX2-ASV: #1 on its trial-set ranking.
Lower-Ranked Scenarios
Scenario ranks below the global position
Show 3 supporting trial sets
- AliMeeting Overlap: #7 on its trial-set ranking.
- VoxCeleb1-O: #7 on its trial-set ranking.
- VOiCES Noise/Reverb: #6 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 |
|---|---|---|---|---|
|
Cross-Lingual Robustness
How well speaker identity survives enrollment-test language mismatch.
|
#1/26
Leader
|
Language mismatch
0.0000 EER from leader
|
1.6478
0.114245 minDCF
|
|
|
Speaking-Style Robustness
Whether the model can preserve identity across emotion-driven change, whispered speech, and noise-induced Lombard speaking style.
|
#1/26
Leader
|
Speaking style shift
0.0000 EER from leader
|
1.7475
0.128603 minDCF
|
|
|
In-The-Wild Robustness
How strong the model is on unconstrained celebrity and media speech across official CN-Celeb and VoxCeleb protocols.
|
#1/26
Leader
|
Open-domain media speech
0.0000 EER from leader
|
2.0151
0.094610 minDCF
|
|
|
Channel / Device Robustness
How resilient the model is to device and channel mismatch.
|
#1/26
Leader
|
Device and channel variation
0.0000 EER from leader
|
2.9828
0.216642 minDCF
|
|
|
Distance Robustness
How much performance changes across meeting and domestic distance mismatch.
|
#1/26
Leader
|
Distance mismatch
0.0000 EER from leader
|
4.0181
0.207658 minDCF
|
|
|
Short-Duration Robustness
How much performance holds up when speech evidence is limited by duration.
|
#1/26
Leader
|
Short-duration speech
0.0000 EER from leader
|
6.3148
0.403125 minDCF
|
|
|
Accent / Dialect Robustness
How reliably the model tracks identity across accent and dialect mismatch.
|
#1/26
Leader
|
Accent and dialect variation
0.0000 EER from leader
|
14.9985
0.476369 minDCF
|
|
|
Genre-Shift Robustness
Whether performance holds when CN-Celeb enrollment and test speech come from different source genres.
|
#2/26
Top 3
|
Source-genre variation
3.0027 EER from leader
|
11.6360
0.658403 minDCF
|
|
|
Aging Robustness
How stable identity representations remain across age-derived and longitudinal recording gaps.
|
#3/26
Top 3
|
Speaker aging and time gaps
0.1192 EER from leader
|
1.9709
0.049742 minDCF
|
|
|
Overlap Robustness
How well speaker identity survives light, mid, and heavy overlap in both meeting and domestic recordings.
|
#5/26
Competitive
|
Overlapping speakers
0.9578 EER from leader
|
8.6494
0.481716 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.
|
#6/26
Competitive
|
Noise and reverberation
0.6000 EER from leader
|
2.8600
0.105900 minDCF
|
Variant Compare
Compare W2V-BERT 2.0 variants
Compare the family variants on the overall board, a Scenario, or a Trial Set.
Show
Variant Compare
Compare W2V-BERT 2.0 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 |
|---|---|---|---|---|---|---|---|---|
|
w2vbert2.0+ MHFA
Current
|
abc | WeSpeaker | DataSet Sampling Share (SpkCount, Hours, NaturalShare) VoxBlink2 25.74% (111,284 spk, 16,122 h, 45.03%) Common Voice 13.07% (67,317 spk, 5,798 h, 22.87%) CN-Celeb 9.86% (2,793 spk, 1,329 h, 2.88%) VoxCeleb2 8.52% (5,994 spk, 2,506 h, 4.96%) TidyVoiceX 8.18% (3,666 spk, 356 h, 1.19%) KeSpeech 7.53% (27,158 spk, 1,076 h, 4.39%) PrivateData 6.03% (19,236 spk, 1,260 h, 2.11%), VOiCES 5.35% (387 spk, 2,947 h, 3.12%), 3D-Speaker 4.51% (10,000 spk, 1,107 h, 2.63%), 3D-Speaker-3A 4.51% (10,000 spk, 3,339 h, 7.90%). | toolkit: wespeaker projection: ArcMarginProduct_intertopk_subcenter_uncertianty DataAug: noise (musan+DNS5_noise) + reverberation(DNS5_RIR+RIRS) | #1 | 5.3492 | 0.267001 | Open |
| w2vbert2.0+ MHFA | pengjy | WeSpeaker | VoxCeleb2 dev + VoxBlink2 + CnCeleb + TidyVoiceX train + 3D-Speaker train + KeSpeech | None | #2 | 5.3542 | 0.228132 | Open |
|
w2v-bert2.0 + cc + ecapa-tdnn + astp + sf2
W2V-BERT 2.0 + MFA
|
JinghanPeng | Contributor-provided | VoxCeleb2 dev + VoxBlink2 + CnCeleb + TidyVoiceX train + 3D-Speaker train + KeSpeech | SphereFace2-C Loss Function + MUSAN/RIRS augmentation | #3 | 6.0847 | 0.235010 | Open |
|
W2V-BERT 2.0 + Adapter-MFA (cross-device consistency FT + weight soup)
W2V-BERT 2.0 + Adapter-MFA
|
ActionPower | WeSpeaker | Init/anchor: public wespeaker W2V-BERT 2.0 + MFA checkpoint (VoxCeleb2 dev + VoxBlink2). Fine-tuning (~40k speakers): VoxCeleb2 dev, CN-Celeb1 dev + CN-Celeb2, 3D-Speaker train, AliMeeting Train (near + far), TidyVoiceX train, Omnilingual ASR corpus, WaxalNLP, and Korean speech corpora (AI-Hub). | LoRA fine-tuning + ArcFace + cross-device consistency loss (parallel multi-device recordings) + augmentation consistency + model soup with the public wespeaker checkpoint | #4 | 7.1907 | 0.288926 | Open |
|
Dual-LoRA W2V-BERT2-Adapter-MFA (TidyVoice)
W2V-BERT 2.0 + Adapter-MFA + Dual-LoRA
|
sgqt | WeSpeaker | VoxCeleb2 dev + VoxBlink2 + TidyVoiceX training split (3,666 speakers, approximately 262k utterances). | Dual-LoRA adaptation of a frozen W2V-BERT2-Adapter-MFA speaker-verification model pre-trained on VoxCeleb2 dev and VoxBlink2. Independent speaker and language LoRA branches are injected across all 24 transformer layers, using ranks 32 and 16, respectively. A shared language-anchored discriminator with gradient reversal removes linguistic information from the speaker representation. Training uses a three-stage curriculum over three epochs, Sub-center ArcMargin loss (K=3, scale=32, margin increased from 0.2 to 0.3), AdamW, 6-second segments, and MUSAN/RIR augmentation. At inference, only the speaker branch is retained and the speaker LoRA weights can be merged into the backbone. | #8 | 9.0606 | 0.342596 | Open |
|
W2V-BERT 2.0 + LoRA_Adapter_MFA (VoxCeleb2 dev + VoxBlink2)
W2V-BERT 2.0 + MFA
|
Public checkpoint | WeSpeaker | VoxCeleb2 dev + VoxBlink2 | W2V-BERT 2.0 speaker-verification recipe with a 24-layer ~600M Conformer frontend and 80-dimensional normalized filterbanks. The MFA backend uses all 25 hidden states with 128-dimensional Layer Adapter compression and a 256-dimensional embedding. Stage 1 freezes the pretrained frontend and trains LoRA adapters on attention query/value projections (rank 64, alpha 128) with ArcFace (margin 0.2, scale 32), 200–300-frame crops, 0.9/1.1 speed perturbation except for VoxBlink2, and MUSAN/RIR augmentation using AdamW with weight decay 1e-4. Stage 2 jointly fine-tunes the frontend and backend for four epochs with cosine decay from 1e-5 to 5e-6. Stage 3 performs large-margin fine-tuning on VoxCeleb2 only without augmentation, using 500–600-frame crops, margin 0.5, and two epochs. AS-Norm and quality-aware score calibration are applied at evaluation. | #10 | 9.9604 | 0.360767 | 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
|
#1/26
|
Cross-lingual
|
200000 |
1.6478
0.114245 minDCF
|
|
HI-MIA
HI-MIA
|
#3/26
|
Short-duration
|
660000 |
2.6133
0.183468 minDCF
|
|
GSC Short
Google Speech Commands
|
#1/26
|
Short-duration
|
220000 |
1.9695
0.148865 minDCF
|
|
GLOBE
GLOBE
|
#1/26
|
Accent/dialect
|
56848 |
23.9164
0.507624 minDCF
|
|
CN-Celeb
CN-Celeb
|
#1/26
|
In-the-wild
|
3484292 |
3.6215
0.168008 minDCF
|
|
CN-Celeb Genre
CN-Celeb
|
#2/26
|
Genre shift
|
440000 |
11.6360
0.658403 minDCF
|
|
CN-Celeb Short
CN-Celeb
|
#2/26
|
Short-duration
|
546964 |
13.1606
0.855317 minDCF
|
|
VoxCeleb1-O
VoxCeleb1
|
#7/26
|
In-the-wild
|
37611 |
0.2339
0.009785 minDCF
|
|
VoxCeleb1-E
VoxCeleb1
|
#4/26
|
In-the-wild
|
579818 |
0.3384
0.017809 minDCF
|
|
VoxCeleb1-H
VoxCeleb1
|
#3/26
|
In-the-wild
|
550894 |
0.6539
0.036041 minDCF
|
|
VoxCeleb Short
VoxCeleb1
|
#1/26
|
Short-duration
|
394724 |
7.5159
0.424850 minDCF
|
|
3D-Speaker Device
3D-Speaker
|
#1/26
|
Channel/device
|
180000 |
2.7433
0.202840 minDCF
|
|
3D-Speaker Distance
3D-Speaker
|
#1/26
|
Distance
|
175163 |
3.0998
0.217095 minDCF
|
|
3D-Speaker Dialect
3D-Speaker
|
#1/26
|
Accent/dialect
|
180000 |
6.0807
0.445113 minDCF
|
|
FFSVC 2022 Cross-Channel
FFSVC 2022
|
#2/26
|
Channel/device
|
72000 |
3.2222
0.230444 minDCF
|
|
FFSVC 2022 Cross-Domain
FFSVC 2022
|
#2/26
|
Distance
|
66546 |
3.2686
0.231439 minDCF
|
|
Whisper40 Whisper
Whisper40
|
#1/26
|
Speaking style
|
17600 |
3.4313
0.273500 minDCF
|
|
Lombard Grid Lombard
Lombard Grid
|
#1/26
|
Speaking style
|
29524 |
0.0037
0.000708 minDCF
|
|
VOiCES Noise/Reverb
VOiCES
|
#6/26
|
Noise/reverb
|
55000 |
2.8600
0.105900 minDCF
|
|
CHiME-6 Domestic Far-Field
CHiME-6
|
#1/26
|
Distance
|
36487 |
5.9391
0.230570 minDCF
|
|
CHiME-6 Overlap
CHiME-6
|
#3/26
|
Overlap
|
39600 |
11.8056
0.664611 minDCF
|
|
ESD
ESD
|
#5/26
|
Speaking style
|
437408 |
1.8075
0.111600 minDCF
|
|
AliMeeting Near/Far
AliMeeting
|
#5/26
|
Distance
|
220000 |
3.7650
0.151530 minDCF
|
|
AliMeeting Overlap
AliMeeting
|
#7/26
|
Overlap
|
165000 |
5.4933
0.298820 minDCF
|
|
VoxKnesset
VoxKnesset
|
#3/26
|
Aging
|
158312 |
2.4333
0.075784 minDCF
|
|
VoxPopuli Aging
VoxPopuli
|
#1/26
|
Aging
|
146575 |
1.5084
0.023700 minDCF
|