System Detail

wespeaker/ res152-voxceleb

wespeaker_resnet152

ranked review: approved WeSpeaker ResNet

This page shows the global rank, scenario ranks, and trial-set results for this submission.

Back To Leaderboard Compare Model Variants Submit Another Run

Global Position

#4/18

Public ranking position among reviewed full-core submissions.

Scenario-Macro EER

11.2275

Scenario-Macro minDCF

0.442748

Coverage

26/26

0 scenarios ranked #1 1 scenario in top 3

Public Listing

Ranked Publicly

This full-core result is approved and included in the public ranking.

Model Provenance

Source and architecture

Displayed name: wespeaker/res152-voxceleb
Source: WeSpeaker
Architecture: ResNet
Submitted alias: wespeaker-voxceleb-resnet152-LM
Created at: 2026-05-29T14:53:46.991718+00:00

Training And Links

Training data and references

Training data: VoxCeleb2 dev, 5,994 speakers
Training setup: ResNet152-TSTP-emb256 r-vector, ArcMargin, 150-epoch VoxCeleb2 recipe with speed perturbation and MUSAN/RIRS augmentation.
Submission mode: full-core
Paper: Deep Residual Learning for Image Recognition

Open Source Checkpoint Open Paper

Higher-Ranked Scenarios

Scenario ranks above the global position

Speaking style #2 In-the-wild #4 Accent/dialect #4

Show 3 supporting trial sets

Lombard Grid Lombard: #3 on its trial-set ranking.
ESD: #3 on its trial-set ranking.
Whisper40 Whisper: #3 on its trial-set ranking.

Lower-Ranked Scenarios

Scenario ranks below the global position

Genre shift #7 Channel/device #7 Distance #7

Show 3 supporting trial sets

3D-Speaker Device: #8 on its trial-set ranking.
AliMeeting Overlap: #8 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
Speaking-Style Robustness Whether the model can preserve identity across emotion-driven change, whispered speech, and noise-induced Lombard speaking style.	#2/18 Top 3	Speaking style shift 1.3434 EER from leader	#3 3.0742 #3 9.2500 #3 0.0745	4.1329 0.268357 minDCF
In-The-Wild Robustness How strong the model is on unconstrained celebrity and media speech across official CN-Celeb and VoxCeleb protocols.	#4/18 Competitive	Open-domain media speech 0.9177 EER from leader	#6 11.3672 #4 0.4998 #6 0.7174 #5 1.2879	6.1011 0.224787 minDCF
Accent / Dialect Robustness How reliably the model tracks identity across accent and dialect mismatch.	#4/18 Competitive	Accent and dialect variation 1.0902 EER from leader	#5 25.0967 #5 17.9540	21.5254 0.675759 minDCF
Aging Robustness How stable identity representations remain across age-derived and longitudinal recording gaps.	#5/18 Competitive	Speaker aging and time gaps 0.6780 EER from leader	#5 3.7467 #5 1.7486	2.7476 0.111015 minDCF
Cross-Lingual Robustness How well speaker identity survives enrollment-test language mismatch.	#5/18 Competitive	Language mismatch 0.3592 EER from leader	#5 4.8269	4.8269 0.305505 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.	#5/18 Competitive	Noise and reverberation 2.6800 EER from leader	#5 5.2600	5.2600 0.154640 minDCF
Overlap Robustness How well speaker identity survives light, mid, and heavy overlap in both meeting and domestic recordings.	#5/18 Competitive	Overlapping speakers 2.8169 EER from leader	#8 9.3467 #5 16.5639	12.9553 0.565796 minDCF
Short-Duration Robustness How much performance holds up when speech evidence is limited by duration.	#5/18 Competitive	Short-duration speech 0.9914 EER from leader	#4 5.8453 #5 10.8860 #5 24.9437 #4 12.3481	13.5058 0.545781 minDCF
Distance Robustness How much performance changes across meeting and domestic distance mismatch.	#7/18 Competitive	Distance mismatch 2.6914 EER from leader	#4 17.5655 #4 8.5600 #5 12.0320 #7 9.6636	11.9552 0.523450 minDCF
Channel / Device Robustness How resilient the model is to device and channel mismatch.	#7/18 Competitive	Device and channel variation 7.9353 EER from leader	#8 18.9800 #7 9.5111	14.2456 0.688961 minDCF
Genre-Shift Robustness Whether performance holds when CN-Celeb enrollment and test speech come from different source genres.	#7/18 Competitive	Source-genre variation 11.9747 EER from leader	#7 26.2472	26.2472 0.806172 minDCF

Variant Compare

Compare ResNet variants

Expand this section to compare checkpoints in the same architecture group by source, training data, training setup, and leaderboard result.

Show

Variant	Source	Training Data	Training Setup	Global Rank	Macro EER	Macro minDCF	Open
wespeaker/res152-voxceleb Current	WeSpeaker	VoxCeleb2 dev, 5,994 speakers	ResNet152-TSTP-emb256 r-vector, ArcMargin, 150-epoch VoxCeleb2 recipe with speed perturbation and MUSAN/RIRS augmentation.	#4 ranked	11.2275	0.442748	Open
wespeaker/res293-voxceleb	WeSpeaker	VoxCeleb2 dev, 5,994 speakers	ResNet293-TSTP-emb256 r-vector, large-margin fine-tuned; official card reports 28.62M parameters and 28.10G FLOPs.	#5 ranked	11.3130	0.439607	Open
wespeaker/res34-voxceleb	WeSpeaker	VoxCeleb2 dev, 5,994 speakers	ResNet34-TSTP-emb256 r-vector, large-margin fine-tuned; official card reports 6.63M parameters and 4.55G FLOPs.	#7 ranked	12.0516	0.477054	Open
wespeaker/res34-cnceleb	WeSpeaker	CN-Celeb train	ResNet34 r-vector with TSTP pooling and large-margin fine-tuning on the CN-Celeb WeSpeaker recipe.	#14 ranked	14.4910	0.607964	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	#5/18	Cross-lingual	200000	4.8269 0.305505 minDCF
HI-MIA HI-MIA	#4/18	Short-duration	660000	5.8453 0.284905 minDCF
GSC Short Google Speech Commands	#5/18	Short-duration	220000	10.8860 0.601260 minDCF
GLOBE GLOBE	#5/18	Accent/dialect	56848	25.0967 0.533224 minDCF
CN-Celeb CN-Celeb	#6/18	In-the-wild	3484292	11.3672 0.399395 minDCF
CN-Celeb Genre CN-Celeb	#7/18	Genre shift	440000	26.2472 0.806172 minDCF
CN-Celeb Short CN-Celeb	#5/18	Short-duration	546964	24.9437 0.851112 minDCF
VoxCeleb1-O VoxCeleb1	#4/18	In-the-wild	37611	0.4998 0.035524 minDCF
VoxCeleb1-E VoxCeleb1	#6/18	In-the-wild	579818	0.7174 0.043120 minDCF
VoxCeleb1-H VoxCeleb1	#5/18	In-the-wild	550894	1.2879 0.071894 minDCF
VoxCeleb Short VoxCeleb1	#4/18	Short-duration	394724	12.3481 0.445848 minDCF
3D-Speaker Device 3D-Speaker	#8/18	Channel/device	180000	18.9800 0.832393 minDCF
3D-Speaker Distance 3D-Speaker	#4/18	Distance	175163	17.5655 0.786467 minDCF
3D-Speaker Dialect 3D-Speaker	#5/18	Accent/dialect	180000	17.9540 0.818293 minDCF
FFSVC 2022 Cross-Channel FFSVC 2022	#7/18	Channel/device	72000	9.5111 0.545528 minDCF
FFSVC 2022 Cross-Domain FFSVC 2022	#7/18	Distance	66546	9.6636 0.556279 minDCF
Whisper40 Whisper Whisper40	#3/18	Speaking style	17600	9.2500 0.567375 minDCF
Lombard Grid Lombard Lombard Grid	#3/18	Speaking style	29524	0.0745 0.011736 minDCF
VOiCES Noise/Reverb VOiCES	#5/18	Noise/reverb	55000	5.2600 0.154640 minDCF
CHiME-6 Domestic Far-Field CHiME-6	#5/18	Distance	36487	12.0320 0.480796 minDCF
CHiME-6 Overlap CHiME-6	#5/18	Overlap	39600	16.5639 0.642111 minDCF
ESD ESD	#3/18	Speaking style	437408	3.0742 0.225959 minDCF
AliMeeting Near/Far AliMeeting	#4/18	Distance	220000	8.5600 0.270260 minDCF
AliMeeting Overlap AliMeeting	#8/18	Overlap	165000	9.3467 0.489480 minDCF
VoxKnesset VoxKnesset	#5/18	Aging	158312	3.7467 0.148597 minDCF
VoxPopuli Aging VoxPopuli	#5/18	Aging	146575	1.7486 0.073433 minDCF