System Detail

W2V-BERT 2.0 + Adapter-MFA (cross-device consistency FT + weight soup)

ActionPower VoicePrint

Submitted by: ActionPower

ranked review: approved WeSpeaker Contributor-provided architecture W2V-BERT 2.0
Global position
#4 / 26
Scenario-macro EER
7.1907
Scenario-macro minDCF
0.288926
Trial-set coverage
26 / 26
Scenario top-three
1
Publicly ranked This result is included in the reviewed public ranking.
Model Provenance

Source and architecture

  • Displayed name: W2V-BERT 2.0 + Adapter-MFA (cross-device consistency FT + weight soup)
  • Submitted by: ActionPower
  • Metadata: Contributor-provided
  • Source: WeSpeaker
  • Contributor-provided architecture: W2V-BERT 2.0 + Adapter-MFA
  • Model family: W2V-BERT 2.0
  • Submitted alias: ActionPower VoicePrint
  • Created at: 2026-11-12T02:53:47.555545+00:00
Training And Links

Training data and references

  • Training data: 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).
  • Training setup: LoRA fine-tuning + ArcFace + cross-device consistency loss (parallel multi-device recordings) + augmentation consistency + model soup with the public wespeaker checkpoint
  • Score processing: None
  • Submission mode: full-core
Higher-Ranked Scenarios

Scenario ranks above the global position

Show 3 supporting trial sets
  • ESD: #2 on its trial-set ranking.
  • VOiCES Noise/Reverb: #2 on its trial-set ranking.
  • VoxCeleb1-O: #4 on its trial-set ranking.
Lower-Ranked Scenarios

Scenario ranks below the global position

Show 3 supporting trial sets
  • Lombard Grid Lombard: #10 on its trial-set ranking.
  • GLOBE: #7 on its trial-set ranking.
  • CN-Celeb Genre: #7 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.
#2/26 Top 3
Noise and reverberation 0.1200 EER from leader
2.3800
0.069900 minDCF
Aging Robustness
How stable identity representations remain across age-derived and longitudinal recording gaps.
#4/26 Competitive
Speaker aging and time gaps 0.1996 EER from leader
2.0513
0.053853 minDCF
Speaking-Style Robustness
Whether the model can preserve identity across emotion-driven change, whispered speech, and noise-induced Lombard speaking style.
#4/26 Competitive
Speaking style shift 0.4225 EER from leader
2.1700
0.129878 minDCF
In-The-Wild Robustness
How strong the model is on unconstrained celebrity and media speech across official CN-Celeb and VoxCeleb protocols.
#4/26 Competitive
Open-domain media speech 1.3188 EER from leader
3.3339
0.136814 minDCF
Overlap Robustness
How well speaker identity survives light, mid, and heavy overlap in both meeting and domestic recordings.
#4/26 Competitive
Overlapping speakers 0.8544 EER from leader
8.5461
0.425573 minDCF
Short-Duration Robustness
How much performance holds up when speech evidence is limited by duration.
#5/26 Competitive
Short-duration speech 2.8557 EER from leader
9.1705
0.365572 minDCF
Cross-Lingual Robustness
How well speaker identity survives enrollment-test language mismatch.
#6/26 Competitive
Language mismatch 0.6395 EER from leader
2.2873
0.179755 minDCF
Distance Robustness
How much performance changes across meeting and domestic distance mismatch.
#6/26 Competitive
Distance mismatch 1.5715 EER from leader
5.5896
0.273671 minDCF
Channel / Device Robustness
How resilient the model is to device and channel mismatch.
#6/26 Competitive
Device and channel variation 2.8092 EER from leader
5.7919
0.363263 minDCF
Accent / Dialect Robustness
How reliably the model tracks identity across accent and dialect mismatch.
#6/26 Competitive
Accent and dialect variation 3.3131 EER from leader
18.3117
0.553127 minDCF
Genre-Shift Robustness
Whether performance holds when CN-Celeb enrollment and test speech come from different source genres.
#7/26 Competitive
Source-genre variation 10.8317 EER from leader
19.4650
0.626777 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 Submitted by Source Training Data Training Setup Rank EER minDCF Open
w2vbert2.0+ MHFA 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
Current
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
Trial Set Rank Scenarios Trials Score
TidyVoiceX2-ASV
TidyVoiceX2-ASV
#6/26
Cross-lingual
200000
2.2873
0.179755 minDCF
HI-MIA
HI-MIA
#6/26
Short-duration
660000
3.4767
0.210408 minDCF
GSC Short
Google Speech Commands
#5/26
Short-duration
220000
4.2485
0.282200 minDCF
GLOBE
GLOBE
#7/26
Accent/dialect
56848
25.0000
0.500793 minDCF
CN-Celeb
CN-Celeb
#7/26
In-the-wild
3484292
6.1954
0.244943 minDCF
CN-Celeb Genre
CN-Celeb
#7/26
Genre shift
440000
19.4650
0.626777 minDCF
CN-Celeb Short
CN-Celeb
#7/26
Short-duration
546964
18.7274
0.601364 minDCF
VoxCeleb1-O
VoxCeleb1
#4/26
In-the-wild
37611
0.2286
0.016646 minDCF
VoxCeleb1-E
VoxCeleb1
#6/26
In-the-wild
579818
0.3842
0.023900 minDCF
VoxCeleb1-H
VoxCeleb1
#6/26
In-the-wild
550894
0.8044
0.045506 minDCF
VoxCeleb Short
VoxCeleb1
#5/26
Short-duration
394724
10.2296
0.368317 minDCF
3D-Speaker Device
3D-Speaker
#7/26
Channel/device
180000
7.2367
0.432887 minDCF
3D-Speaker Distance
3D-Speaker
#5/26
Distance
175163
6.0867
0.363766 minDCF
3D-Speaker Dialect
3D-Speaker
#6/26
Accent/dialect
180000
11.6233
0.605460 minDCF
FFSVC 2022 Cross-Channel
FFSVC 2022
#6/26
Channel/device
72000
4.3472
0.293639 minDCF
FFSVC 2022 Cross-Domain
FFSVC 2022
#6/26
Distance
66546
4.4428
0.299977 minDCF
Whisper40 Whisper
Whisper40
#5/26
Speaking style
17600
4.9938
0.275938 minDCF
Lombard Grid Lombard
Lombard Grid
#10/26
Speaking style
29524
0.0633
0.005067 minDCF
VOiCES Noise/Reverb
VOiCES
#2/26
Noise/reverb
55000
2.3800
0.069900 minDCF
CHiME-6 Domestic Far-Field
CHiME-6
#4/26
Distance
36487
6.9340
0.268405 minDCF
CHiME-6 Overlap
CHiME-6
#5/26
Overlap
39600
11.9361
0.599472 minDCF
ESD
ESD
#2/26
Speaking style
437408
1.4529
0.108628 minDCF
AliMeeting Near/Far
AliMeeting
#7/26
Distance
220000
4.8950
0.162535 minDCF
AliMeeting Overlap
AliMeeting
#6/26
Overlap
165000
5.1560
0.251673 minDCF
VoxKnesset
VoxKnesset
#4/26
Aging
158312
2.5867
0.082648 minDCF
VoxPopuli Aging
VoxPopuli
#4/26
Aging
146575
1.5159
0.025058 minDCF