System Detail

W2V-BERT 2.0 + MHFA (VoxCeleb2 + VoxBlink2)

Wav2Vec2-BERT

Submitted by: pengjy

ranked review: approved WeSpeaker W2V-BERT 2.0 + MHFA W2V-BERT 2.0

Global position: #1 / 23
Scenario-macro EER: 7.5635
Scenario-macro minDCF: 0.368244
Trial-set coverage: 26 / 26
Scenario top-three: 9

Publicly ranked This result is included in the reviewed public ranking.

Global leaderboard Compare variants Submit another run

Model Provenance

Source and architecture

Displayed name: W2V-BERT 2.0 + MHFA (VoxCeleb2 + VoxBlink2)
Submitted by: pengjy
Metadata: OpenSVBench model record
Source: WeSpeaker
Architecture: W2V-BERT 2.0 + MHFA
Model family: W2V-BERT 2.0
Submitted alias: BUT/w2v-bert-2.0-mhfa-vox2-vb2
Created at: 2026-10-18T20:56:28.807794+00:00

Training And Links

Training data and references

Training data: VoxCeleb2 dev + VoxBlink2: 111,301 speakers, 4.83M utterances.
Training setup: Frontend: facebook/w2v-bert-2.0 (600M). Backend: MHFA with LayerNorm (64 heads, 128-d compression, all 25 hidden states), 256-d embedding. Loss: sub-center ArcFace (K=3, s=32).
Submission mode: full-core
Model source: Open link
Paper: An Attention-Based Backend Allowing Efficient Fine-Tuning of Transformer Models for Speaker Verification

Open Model Source Open Paper

Higher-Ranked Scenarios

Scenario ranks above the global position

Aging #1 Speaking style #1 Overlap #1

Show 3 supporting trial sets

VoxKnesset: #1 on its trial-set ranking.
Whisper40 Whisper: #1 on its trial-set ranking.
CHiME-6 Overlap: #1 on its trial-set ranking.

Lower-Ranked Scenarios

Scenario ranks below the global position

Genre shift #4 Cross-lingual #4 Channel/device #3

Show 3 supporting trial sets

GSC Short: #6 on its trial-set ranking.
VoxCeleb1-O: #5 on its trial-set ranking.
CN-Celeb Genre: #4 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.	#1/23 Leader	Speaker aging and time gaps 0.0000 EER from leader	#1 2.0600 #4 1.5235	1.7917 0.060666 minDCF
Speaking-Style Robustness Whether the model can preserve identity across emotion-driven change, whispered speech, and noise-induced Lombard speaking style.	#1/23 Leader	Speaking style shift 0.0000 EER from leader	#3 1.7975 #1 3.9375 #2 0.0149	1.9166 0.153402 minDCF
Overlap Robustness How well speaker identity survives light, mid, and heavy overlap in both meeting and domestic recordings.	#1/23 Leader	Overlapping speakers 0.0000 EER from leader	#3 5.1047 #1 11.1139	8.1093 0.560809 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.	#2/23 Top 3	Noise and reverberation 0.1400 EER from leader	#2 2.4000	2.4000 0.086400 minDCF
In-The-Wild Robustness How strong the model is on unconstrained celebrity and media speech across official CN-Celeb and VoxCeleb protocols.	#2/23 Top 3	Open-domain media speech 0.0493 EER from leader	#3 7.0671 #5 0.2658 #4 0.4005 #4 0.7583	3.7710 0.155142 minDCF
Distance Robustness How much performance changes across meeting and domestic distance mismatch.	#2/23 Top 3	Distance mismatch 1.5592 EER from leader	#2 9.6287 #3 5.4400 #2 6.5421 #2 5.6533	6.8160 0.373691 minDCF
Short-Duration Robustness How much performance holds up when speech evidence is limited by duration.	#2/23 Top 3	Short-duration speech 0.3063 EER from leader	#3 3.6067 #6 7.0645 #1 17.5046 #2 9.1879	9.3409 0.524054 minDCF
Accent / Dialect Robustness How reliably the model tracks identity across accent and dialect mismatch.	#2/23 Top 3	Accent and dialect variation 2.3627 EER from leader	#1 23.6455 #3 14.7233	19.1844 0.677736 minDCF
Channel / Device Robustness How resilient the model is to device and channel mismatch.	#3/23 Top 3	Device and channel variation 4.2054 EER from leader	#3 10.8460 #2 5.5694	8.2077 0.509507 minDCF
Cross-Lingual Robustness How well speaker identity survives enrollment-test language mismatch.	#4/23 Competitive	Language mismatch 1.0028 EER from leader	#4 2.6605	2.6605 0.211953 minDCF
Genre-Shift Robustness Whether performance holds when CN-Celeb enrollment and test speech come from different source genres.	#4/23 Competitive	Source-genre variation 4.7280 EER from leader	#4 19.0005	19.0005 0.737320 minDCF

Variant Compare

Compare W2V-BERT 2.0 variants

Expand this section to compare models in the same model family by architecture, source, training data, training setup, and leaderboard result.

Show

Variant	Submitted by	Source	Training Data	Training Setup	Global Rank	Macro EER	Macro minDCF	Open
W2V-BERT 2.0 + MHFA (VoxCeleb2 + VoxBlink2) W2V-BERT 2.0 + MHFA Current	pengjy	WeSpeaker	VoxCeleb2 dev + VoxBlink2: 111,301 speakers, 4.83M utterances.	Frontend: facebook/w2v-bert-2.0 (600M). Backend: MHFA with LayerNorm (64 heads, 128-d compression, all 25 hidden states), 256-d embedding. Loss: sub-center ArcFace (K=3, s=32).	#1 ranked	7.5635	0.368244	Open
Dual-LoRA W2V-BERT2-Adapter-MFA (TidyVoice) W2V-BERT 2.0 + Adapter-MFA + Dual-LoRA	shangguanqituan	WeSpeaker	VoxBlink2 and 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 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.	#4 ranked	9.0606	0.342596	Open
wespeaker/w2vbert2 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.	#6 ranked	9.9604	0.360767	Open

Trial-Set Results

Raw ranking by trial set

Expand this section to inspect the detailed trial-set rankings.

Show

Trial Set	Rank	Scenarios	Trials	Score
TidyVoiceX2-ASV TidyVoiceX2-ASV	#4/23	Cross-lingual	200000	2.6605 0.211953 minDCF
HI-MIA HI-MIA	#3/23	Short-duration	660000	3.6067 0.210263 minDCF
GSC Short Google Speech Commands	#6/23	Short-duration	220000	7.0645 0.517445 minDCF
GLOBE GLOBE	#1/23	Accent/dialect	56848	23.6455 0.513913 minDCF
CN-Celeb CN-Celeb	#3/23	In-the-wild	3484292	7.0671 0.277016 minDCF
CN-Celeb Genre CN-Celeb	#4/23	Genre shift	440000	19.0005 0.737320 minDCF
CN-Celeb Short CN-Celeb	#1/23	Short-duration	546964	17.5046 0.797239 minDCF
VoxCeleb1-O VoxCeleb1	#5/23	In-the-wild	37611	0.2658 0.025633 minDCF
VoxCeleb1-E VoxCeleb1	#4/23	In-the-wild	579818	0.4005 0.026122 minDCF
VoxCeleb1-H VoxCeleb1	#4/23	In-the-wild	550894	0.7583 0.048047 minDCF
VoxCeleb Short VoxCeleb1	#2/23	Short-duration	394724	9.1879 0.571269 minDCF
3D-Speaker Device 3D-Speaker	#3/23	Channel/device	180000	10.8460 0.643653 minDCF
3D-Speaker Distance 3D-Speaker	#2/23	Distance	175163	9.6287 0.589211 minDCF
3D-Speaker Dialect 3D-Speaker	#3/23	Accent/dialect	180000	14.7233 0.841560 minDCF
FFSVC 2022 Cross-Channel FFSVC 2022	#2/23	Channel/device	72000	5.5694 0.375361 minDCF
FFSVC 2022 Cross-Domain FFSVC 2022	#2/23	Distance	66546	5.6533 0.381613 minDCF
Whisper40 Whisper Whisper40	#1/23	Speaking style	17600	3.9375 0.310375 minDCF
Lombard Grid Lombard Lombard Grid	#2/23	Speaking style	29524	0.0149 0.001118 minDCF
VOiCES Noise/Reverb VOiCES	#2/23	Noise/reverb	55000	2.4000 0.086400 minDCF
CHiME-6 Domestic Far-Field CHiME-6	#2/23	Distance	36487	6.5421 0.343925 minDCF
CHiME-6 Overlap CHiME-6	#1/23	Overlap	39600	11.1139 0.755278 minDCF
ESD ESD	#3/23	Speaking style	437408	1.7975 0.148715 minDCF
AliMeeting Near/Far AliMeeting	#3/23	Distance	220000	5.4400 0.180015 minDCF
AliMeeting Overlap AliMeeting	#3/23	Overlap	165000	5.1047 0.366340 minDCF
VoxKnesset VoxKnesset	#1/23	Aging	158312	2.0600 0.091088 minDCF
VoxPopuli Aging VoxPopuli	#4/23	Aging	146575	1.5235 0.030244 minDCF