System Detail
palabraai/redimnet2-b6-vb2+vox2
redimnet2-b6-vb2vox2-lm-full-core
Submitted by: Public checkpoint
- Global position
- #8 / 25
- Scenario-macro EER
- 9.1442
- Scenario-macro minDCF
- 0.324160
- Trial-set coverage
- 26 / 26
- Scenario top-three
- 2
Publicly ranked
This result is included in the reviewed public ranking.
Model Provenance
Source and architecture
- Displayed name:
palabraai/redimnet2-b6-vb2+vox2 - Submitted by: Public checkpoint
- Metadata: OpenSVBench model record
- Source: PalabraAI
- Architecture: ReDimNet2-B6
- Model family: ReDimNet2
- Submitted alias:
b6-vb2+vox2_v0-lm.pt - Created at:
2026-10-16T09:04:30.545942+00:00
Training And Links
Training data and references
- Training data: VoxCeleb2 dev + VoxBlink2
- Training setup: ReDimNet2-B6 speaker-verification recipe: 80-dimensional mean-normalized log-Mel filterbanks (25 ms window, 10 ms shift), a time-pooled dimension-reshaping backbone with 192-dimensional embeddings, 2-second random crops, MUSAN/RIR augmentation, and 0.9/1.1 speed perturbation during SphereFace2-C training with margin warm-up. Large-margin fine-tuning uses 6-second segments without speed perturbation, fixed margin 0.3, and five epochs with exponential learning-rate decay. This checkpoint is the extended VoxCeleb2 + VoxBlink2 release; the public release reports 12.3M parameters and 13.05 GMACs.
- Score processing: None
- Submission mode:
full-core - Model source: Open link
- Paper: ReDimNet2: Scaling Speaker Verification via Time-Pooled Dimension Reshaping
Higher-Ranked Scenarios
Scenario ranks above the global position
Show 3 supporting trial sets
- VOiCES Noise/Reverb: #1 on its trial-set ranking.
- VoxCeleb1-O: #3 on its trial-set ranking.
- VoxCeleb1-H: #3 on its trial-set ranking.
Lower-Ranked Scenarios
Scenario ranks below the global position
Show 3 supporting trial sets
- GLOBE: #14 on its trial-set ranking.
- CN-Celeb Short: #13 on its trial-set ranking.
- CN-Celeb Genre: #11 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.
|
#1/25
Leader
|
Noise and reverberation
0.0000 EER from leader
|
2.2600
0.073780 minDCF
|
|
|
Aging Robustness
How stable identity representations remain across age-derived and longitudinal recording gaps.
|
#3/25
Top 3
|
Speaker aging and time gaps
0.1129 EER from leader
|
1.9046
0.052327 minDCF
|
|
|
Speaking-Style Robustness
Whether the model can preserve identity across emotion-driven change, whispered speech, and noise-induced Lombard speaking style.
|
#5/25
Competitive
|
Speaking style shift
0.4625 EER from leader
|
2.3792
0.145230 minDCF
|
|
|
Overlap Robustness
How well speaker identity survives light, mid, and heavy overlap in both meeting and domestic recordings.
|
#5/25
Competitive
|
Overlapping speakers
1.3028 EER from leader
|
8.9944
0.440323 minDCF
|
|
|
In-The-Wild Robustness
How strong the model is on unconstrained celebrity and media speech across official CN-Celeb and VoxCeleb protocols.
|
#6/25
Competitive
|
Open-domain media speech
1.7263 EER from leader
|
4.1123
0.141370 minDCF
|
|
|
Cross-Lingual Robustness
How well speaker identity survives enrollment-test language mismatch.
|
#8/25
Competitive
|
Language mismatch
1.7614 EER from leader
|
3.4191
0.259714 minDCF
|
|
|
Distance Robustness
How much performance changes across meeting and domestic distance mismatch.
|
#8/25
Competitive
|
Distance mismatch
4.8416 EER from leader
|
9.5358
0.334705 minDCF
|
|
|
Channel / Device Robustness
How resilient the model is to device and channel mismatch.
|
#8/25
Competitive
|
Device and channel variation
7.1347 EER from leader
|
11.1008
0.464227 minDCF
|
|
|
Short-Duration Robustness
How much performance holds up when speech evidence is limited by duration.
|
#8/25
Competitive
|
Short-duration speech
3.5820 EER from leader
|
12.1765
0.431643 minDCF
|
|
|
Accent / Dialect Robustness
How reliably the model tracks identity across accent and dialect mismatch.
|
#11/25
Competitive
|
Accent and dialect variation
5.4176 EER from leader
|
21.1791
0.581639 minDCF
|
|
|
Genre-Shift Robustness
Whether performance holds when CN-Celeb enrollment and test speech come from different source genres.
|
#11/25
Competitive
|
Source-genre variation
10.2845 EER from leader
|
23.5245
0.640800 minDCF
|
Variant Compare
Compare ReDimNet2 variants
Compare the family variants on the overall board, a Scenario, or a Trial Set.
Show
Variant Compare
Compare ReDimNet2 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 |
|---|---|---|---|---|---|---|---|---|
|
palabraai/redimnet2-b6-vb2+vox2
ReDimNet2-B6
Current
|
Public checkpoint | PalabraAI | VoxCeleb2 dev + VoxBlink2 | ReDimNet2-B6 speaker-verification recipe: 80-dimensional mean-normalized log-Mel filterbanks (25 ms window, 10 ms shift), a time-pooled dimension-reshaping backbone with 192-dimensional embeddings, 2-second random crops, MUSAN/RIR augmentation, and 0.9/1.1 speed perturbation during SphereFace2-C training with margin warm-up. Large-margin fine-tuning uses 6-second segments without speed perturbation, fixed margin 0.3, and five epochs with exponential learning-rate decay. This checkpoint is the extended VoxCeleb2 + VoxBlink2 release; the public release reports 12.3M parameters and 13.05 GMACs. | #8 | 9.1442 | 0.324160 | Open |
|
palabraai/redimnet2-b6
ReDimNet2-B6
|
Public checkpoint | PalabraAI | VoxCeleb2 dev | ReDimNet2-B6 speaker-verification recipe: 80-dimensional mean-normalized log-Mel filterbanks (25 ms window, 10 ms shift), a time-pooled dimension-reshaping backbone with 192-dimensional embeddings, 2-second random crops, MUSAN/RIR augmentation, and 0.9/1.1 speed perturbation during SphereFace2-C training with margin warm-up. Large-margin fine-tuning uses 6-second segments without speed perturbation, fixed margin 0.3, and five epochs with exponential learning-rate decay. The public release reports 12.3M parameters and 13.05 GMACs. | #10 | 10.5277 | 0.419432 | 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
|
#8/25
|
Cross-lingual
|
200000 |
3.4191
0.259714 minDCF
|
|
HI-MIA
HI-MIA
|
#11/25
|
Short-duration
|
660000 |
5.8667
0.240098 minDCF
|
|
GSC Short
Google Speech Commands
|
#10/25
|
Short-duration
|
220000 |
8.1985
0.487250 minDCF
|
|
GLOBE
GLOBE
|
#14/25
|
Accent/dialect
|
56848 |
25.2515
0.507624 minDCF
|
|
CN-Celeb
CN-Celeb
|
#9/25
|
In-the-wild
|
3484292 |
7.8288
0.259349 minDCF
|
|
CN-Celeb Genre
CN-Celeb
|
#11/25
|
Genre shift
|
440000 |
23.5245
0.640800 minDCF
|
|
CN-Celeb Short
CN-Celeb
|
#13/25
|
Short-duration
|
546964 |
25.1046
0.644055 minDCF
|
|
VoxCeleb1-O
VoxCeleb1
|
#3/25
|
In-the-wild
|
37611 |
0.1701
0.011700 minDCF
|
|
VoxCeleb1-E
VoxCeleb1
|
#4/25
|
In-the-wild
|
579818 |
0.3387
0.019785 minDCF
|
|
VoxCeleb1-H
VoxCeleb1
|
#3/25
|
In-the-wild
|
550894 |
0.6788
0.038685 minDCF
|
|
VoxCeleb Short
VoxCeleb1
|
#3/25
|
Short-duration
|
394724 |
9.5363
0.355169 minDCF
|
|
3D-Speaker Device
3D-Speaker
|
#10/25
|
Channel/device
|
180000 |
15.3933
0.565787 minDCF
|
|
3D-Speaker Distance
3D-Speaker
|
#10/25
|
Distance
|
175163 |
15.4780
0.513483 minDCF
|
|
3D-Speaker Dialect
3D-Speaker
|
#9/25
|
Accent/dialect
|
180000 |
17.1067
0.655653 minDCF
|
|
FFSVC 2022 Cross-Channel
FFSVC 2022
|
#9/25
|
Channel/device
|
72000 |
6.8083
0.362667 minDCF
|
|
FFSVC 2022 Cross-Domain
FFSVC 2022
|
#9/25
|
Distance
|
66546 |
6.9360
0.373640 minDCF
|
|
Whisper40 Whisper
Whisper40
|
#3/25
|
Speaking style
|
17600 |
4.7875
0.269813 minDCF
|
|
Lombard Grid Lombard
Lombard Grid
|
#4/25
|
Speaking style
|
29524 |
0.0373
0.001453 minDCF
|
|
VOiCES Noise/Reverb
VOiCES
|
#1/25
|
Noise/reverb
|
55000 |
2.2600
0.073780 minDCF
|
|
CHiME-6 Domestic Far-Field
CHiME-6
|
#6/25
|
Distance
|
36487 |
7.3741
0.266385 minDCF
|
|
CHiME-6 Overlap
CHiME-6
|
#4/25
|
Overlap
|
39600 |
11.8889
0.584639 minDCF
|
|
ESD
ESD
|
#7/25
|
Speaking style
|
437408 |
2.3127
0.164424 minDCF
|
|
AliMeeting Near/Far
AliMeeting
|
#11/25
|
Distance
|
220000 |
8.3550
0.185310 minDCF
|
|
AliMeeting Overlap
AliMeeting
|
#7/25
|
Overlap
|
165000 |
6.1000
0.296007 minDCF
|
|
VoxKnesset
VoxKnesset
|
#3/25
|
Aging
|
158312 |
2.2933
0.078447 minDCF
|
|
VoxPopuli Aging
VoxPopuli
|
#4/25
|
Aging
|
146575 |
1.5159
0.026206 minDCF
|