System Detail

AntSpeaker/MECT-B2

antspeaker-mect-b2-vb2-full-core

Submitted by: Public checkpoint

ranked review: approved AntSpeaker MECT-B2 MECT

Global position: #6 / 25
Scenario-macro EER: 8.5599
Scenario-macro minDCF: 0.321518
Trial-set coverage: 26 / 26
Scenario top-three: 1

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: AntSpeaker/MECT-B2
Submitted by: Public checkpoint
Metadata: OpenSVBench model record
Source: AntSpeaker
Architecture: MECT-B2
Model family: MECT
Submitted alias: AntSpeaker/MECT-B2
Created at: 2026-10-19T05:05:41.317553+00:00

Training And Links

Training data and references

Training data: VoxCeleb2 dev + VoxBlink2
Training setup: MECT-B2 uses 80-dimensional mean-normalized log-Mel filterbanks at 16 kHz with a 25 ms window, 10 ms shift, and 20–7600 Hz band, followed by a four-stage CNN-Transformer with C=32, M=3, and Transformer width d=64. The model uses frame-level dense mixture-of-experts with four experts, weighted dense fusion, attentive statistics pooling, and a 192-dimensional projection trained with SphereFace2. The VoxCeleb2 + VoxBlink2 setting uses full training followed by large-margin fine-tuning without speed perturbation; the LMF stage uses VoxCeleb2.
Score processing: None
Submission mode: full-core
Model source: Open link
Paper: MECT: Mixture of Experts with CNN-Transformer Network for Speaker Verification

Open Model Source Open Paper

Higher-Ranked Scenarios

Scenario ranks above the global position

Aging #2 In-the-wild #4 Noise/reverb #5

Show 3 supporting trial sets

Lombard Grid Lombard: #1 on its trial-set ranking.
VoxCeleb1-E: #1 on its trial-set ranking.
VoxCeleb1-H: #1 on its trial-set ranking.

Lower-Ranked Scenarios

Scenario ranks below the global position

Genre shift #10 Channel/device #7 Speaking style #7

Show 3 supporting trial sets

GSC Short: #12 on its trial-set ranking.
CN-Celeb Genre: #10 on its trial-set ranking.
CN-Celeb Short: #10 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.	#2/25 Top 3	Speaker aging and time gaps 0.0600 EER from leader	#2 2.1800 #5 1.5235	1.8517 0.056173 minDCF
In-The-Wild Robustness How strong the model is on unconstrained celebrity and media speech across official CN-Celeb and VoxCeleb protocols.	#4/25 Competitive	Open-domain media speech 1.3357 EER from leader	#7 7.0910 #5 0.2287 #1 0.2901 #1 0.5387	3.7217 0.130256 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.	#5/25 Competitive	Noise and reverberation 0.3940 EER from leader	#5 2.6540	2.6540 0.087120 minDCF
Cross-Lingual Robustness How well speaker identity survives enrollment-test language mismatch.	#6/25 Competitive	Language mismatch 0.9634 EER from leader	#6 2.6211	2.6211 0.254113 minDCF
Distance Robustness How much performance changes across meeting and domestic distance mismatch.	#6/25 Competitive	Distance mismatch 3.3564 EER from leader	#7 12.3634 #10 7.7800 #1 6.2828 #6 5.7764	8.0506 0.332292 minDCF
Overlap Robustness How well speaker identity survives light, mid, and heavy overlap in both meeting and domestic recordings.	#6/25 Competitive	Overlapping speakers 1.3144 EER from leader	#8 6.5733 #3 11.4389	9.0061 0.442528 minDCF
Short-Duration Robustness How much performance holds up when speech evidence is limited by duration.	#6/25 Competitive	Short-duration speech 2.2174 EER from leader	#8 3.7550 #12 8.8100 #10 21.5288 #1 9.1539	10.8119 0.429590 minDCF
Accent / Dialect Robustness How reliably the model tracks identity across accent and dialect mismatch.	#6/25 Competitive	Accent and dialect variation 4.2294 EER from leader	#9 25.0484 #7 14.9333	19.9909 0.569899 minDCF
Speaking-Style Robustness Whether the model can preserve identity across emotion-driven change, whispered speech, and noise-induced Lombard speaking style.	#7/25 Competitive	Speaking style shift 1.0292 EER from leader	#1 1.2675 #8 7.5625 #1 0.0075	2.9458 0.150257 minDCF
Channel / Device Robustness How resilient the model is to device and channel mismatch.	#7/25 Competitive	Device and channel variation 5.1260 EER from leader	#7 12.5620 #6 5.6222	9.0921 0.435914 minDCF
Genre-Shift Robustness Whether performance holds when CN-Celeb enrollment and test speech come from different source genres.	#10/25 Competitive	Source-genre variation 10.1725 EER from leader	#10 23.4125	23.4125 0.648552 minDCF

Variant Compare

Compare MECT variants

Compare the family variants on the overall board, a Scenario, or a Trial Set.

Compare by

Show

Variant	Submitted by	Source	Training Data	Training Setup	Rank	EER	minDCF	Open
MECT-b2-czur-finetuned MECT-B2	czur	mect	VoxCeleb2 dev + VoxBlink2 + 3D-speaker	Follow AntSpeaker: MECT-B2 adopts 80-dim mean-normalized log-Mel filterbanks (16 kHz, 25 ms window, 10 ms hop, 20–7600 Hz). Its 4-stage CNN-Transformer has C=32, M=3, d=64. It employs frame-level 4-expert MoE, weighted dense fusion, attentive stats pooling and 192-dim projection, trained via SphereFace2. It is further fine-tuned on 3D-Speaker to optimize for dialect variations and varying recording distances.	#3	7.2276	0.280633	Open
AntSpeaker/MECT-B2 MECT-B2 Current	Public checkpoint	AntSpeaker	VoxCeleb2 dev + VoxBlink2	MECT-B2 uses 80-dimensional mean-normalized log-Mel filterbanks at 16 kHz with a 25 ms window, 10 ms shift, and 20–7600 Hz band, followed by a four-stage CNN-Transformer with C=32, M=3, and Transformer width d=64. The model uses frame-level dense mixture-of-experts with four experts, weighted dense fusion, attentive statistics pooling, and a 192-dimensional projection trained with SphereFace2. The VoxCeleb2 + VoxBlink2 setting uses full training followed by large-margin fine-tuning without speed perturbation; the LMF stage uses VoxCeleb2.	#6	8.5599	0.321518	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	#6/25	Cross-lingual	200000	2.6211 0.254113 minDCF
HI-MIA HI-MIA	#8/25	Short-duration	660000	3.7550 0.188360 minDCF
GSC Short Google Speech Commands	#12/25	Short-duration	220000	8.8100 0.492470 minDCF
GLOBE GLOBE	#9/25	Accent/dialect	56848	25.0484 0.505824 minDCF
CN-Celeb CN-Celeb	#7/25	In-the-wild	3484292	7.0910 0.240625 minDCF
CN-Celeb Genre CN-Celeb	#10/25	Genre shift	440000	23.4125 0.648552 minDCF
CN-Celeb Short CN-Celeb	#10/25	Short-duration	546964	21.5288 0.637650 minDCF
VoxCeleb1-O VoxCeleb1	#5/25	In-the-wild	37611	0.2287 0.011699 minDCF
VoxCeleb1-E VoxCeleb1	#1/25	In-the-wild	579818	0.2901 0.016274 minDCF
VoxCeleb1-H VoxCeleb1	#1/25	In-the-wild	550894	0.5387 0.031688 minDCF
VoxCeleb Short VoxCeleb1	#1/25	Short-duration	394724	9.1539 0.399880 minDCF
3D-Speaker Device 3D-Speaker	#7/25	Channel/device	180000	12.5620 0.544773 minDCF
3D-Speaker Distance 3D-Speaker	#7/25	Distance	175163	12.3634 0.528588 minDCF
3D-Speaker Dialect 3D-Speaker	#7/25	Accent/dialect	180000	14.9333 0.633973 minDCF
FFSVC 2022 Cross-Channel FFSVC 2022	#6/25	Channel/device	72000	5.6222 0.327056 minDCF
FFSVC 2022 Cross-Domain FFSVC 2022	#6/25	Distance	66546	5.7764 0.333581 minDCF
Whisper40 Whisper Whisper40	#8/25	Speaking style	17600	7.5625 0.354313 minDCF
Lombard Grid Lombard Lombard Grid	#1/25	Speaking style	29524	0.0075 0.000373 minDCF
VOiCES Noise/Reverb VOiCES	#5/25	Noise/reverb	55000	2.6540 0.087120 minDCF
CHiME-6 Domestic Far-Field CHiME-6	#1/25	Distance	36487	6.2828 0.277510 minDCF
CHiME-6 Overlap CHiME-6	#3/25	Overlap	39600	11.4389 0.570083 minDCF
ESD ESD	#1/25	Speaking style	437408	1.2675 0.096087 minDCF
AliMeeting Near/Far AliMeeting	#10/25	Distance	220000	7.7800 0.189490 minDCF
AliMeeting Overlap AliMeeting	#8/25	Overlap	165000	6.5733 0.314973 minDCF
VoxKnesset VoxKnesset	#2/25	Aging	158312	2.1800 0.086290 minDCF
VoxPopuli Aging VoxPopuli	#5/25	Aging	146575	1.5235 0.026056 minDCF