System Detail

microsoft/ unispeech-sat-base-plus-sv

unispeech_sat_base_plus

ranked review: approved Microsoft UniSpeech-SAT Base+

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

Back To Leaderboard Submit Another Run

Global Position

#18/18

Public ranking position among reviewed full-core submissions.

Scenario-Macro EER

25.6711

Scenario-Macro minDCF

0.890736

Coverage

26/26

0 scenarios ranked #1 0 scenarios 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: microsoft/unispeech-sat-base-plus-sv
Source: Microsoft
Architecture: UniSpeech-SAT Base+
Submitted alias: unispeech-sat-base-plus-sv
Created at: 2026-05-31T02:57:39.101571+00:00

Training And Links

Training data and references

Training data: Pretrained on 60k h Libri-Light, 10k h GigaSpeech, and 24k h VoxPopuli; SV head fine-tuned on VoxCeleb1
Training setup: UniSpeech-SAT Base+ speaker-aware SSL model fine-tuned for speaker verification with an x-vector head and additive-margin softmax.
Submission mode: full-core
Paper: UniSpeech-SAT: Universal Speech Representation Learning with Speaker Aware Pre-Training

Open Source Checkpoint Open Paper

Higher-Ranked Scenarios

Scenario ranks above the global position

Noise/reverb #16 Short-duration #17 Distance #17

Show 3 supporting trial sets

VoxCeleb Short: #12 on its trial-set ranking.
VoxCeleb1-E: #13 on its trial-set ranking.
VoxCeleb1-O: #13 on its trial-set ranking.

Lower-Ranked Scenarios

Scenario ranks below the global position

Channel/device #18 Overlap #18 Cross-lingual #18

Show 3 supporting trial sets

Whisper40 Whisper: #18 on its trial-set ranking.
CN-Celeb: #18 on its trial-set ranking.
VoxKnesset: #18 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.	#16/18 Needs work	Noise and reverberation 18.4000 EER from leader	#16 20.9800	20.9800 0.741000 minDCF
Short-Duration Robustness How much performance holds up when speech evidence is limited by duration.	#17/18 Needs work	Short-duration speech 13.6334 EER from leader	#17 23.7600 #17 24.4600 #17 34.0117 #12 22.3595	26.1478 0.927872 minDCF
Distance Robustness How much performance changes across meeting and domestic distance mismatch.	#17/18 Needs work	Distance mismatch 19.5817 EER from leader	#17 37.8673 #17 23.7800 #17 26.7802 #18 26.9545	28.8455 0.982128 minDCF
Accent / Dialect Robustness How reliably the model tracks identity across accent and dialect mismatch.	#17/18 Needs work	Accent and dialect variation 12.1707 EER from leader	#14 26.9350 #17 38.2767	32.6058 0.912056 minDCF
Genre-Shift Robustness Whether performance holds when CN-Celeb enrollment and test speech come from different source genres.	#17/18 Needs work	Source-genre variation 22.7618 EER from leader	#17 37.0343	37.0343 0.985940 minDCF
In-The-Wild Robustness How strong the model is on unconstrained celebrity and media speech across official CN-Celeb and VoxCeleb protocols.	#18/18 Needs work	Open-domain media speech 12.5012 EER from leader	#18 29.8022 #13 5.6037 #13 3.6382 #13 7.4593	17.6846 0.670598 minDCF
Speaking-Style Robustness Whether the model can preserve identity across emotion-driven change, whispered speech, and noise-induced Lombard speaking style.	#18/18 Needs work	Speaking style shift 15.4003 EER from leader	#16 12.8825 #18 33.6875 #17 7.9993	18.1898 0.805772 minDCF
Aging Robustness How stable identity representations remain across age-derived and longitudinal recording gaps.	#18/18 Needs work	Speaker aging and time gaps 17.3082 EER from leader	#18 27.7733 #17 10.9824	19.3778 0.858414 minDCF
Cross-Lingual Robustness How well speaker identity survives enrollment-test language mismatch.	#18/18 Needs work	Language mismatch 18.7612 EER from leader	#18 23.2288	23.2288 0.970581 minDCF
Overlap Robustness How well speaker identity survives light, mid, and heavy overlap in both meeting and domestic recordings.	#18/18 Needs work	Overlapping speakers 16.3639 EER from leader	#18 23.5933 #16 29.4111	26.5022 0.944448 minDCF
Channel / Device Robustness How resilient the model is to device and channel mismatch.	#18/18 Needs work	Device and channel variation 25.4754 EER from leader	#17 36.8213 #18 26.7500	31.7857 0.999283 minDCF

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	#18/18	Cross-lingual	200000	23.2288 0.970581 minDCF
HI-MIA HI-MIA	#17/18	Short-duration	660000	23.7600 0.944420 minDCF
GSC Short Google Speech Commands	#17/18	Short-duration	220000	24.4600 0.966790 minDCF
GLOBE GLOBE	#14/18	Accent/dialect	56848	26.9350 0.824245 minDCF
CN-Celeb CN-Celeb	#18/18	In-the-wild	3484292	29.8022 0.945410 minDCF
CN-Celeb Genre CN-Celeb	#17/18	Genre shift	440000	37.0343 0.985940 minDCF
CN-Celeb Short CN-Celeb	#17/18	Short-duration	546964	34.0117 0.998459 minDCF
VoxCeleb1-O VoxCeleb1	#13/18	In-the-wild	37611	5.6037 0.466737 minDCF
VoxCeleb1-E VoxCeleb1	#13/18	In-the-wild	579818	3.6382 0.261858 minDCF
VoxCeleb1-H VoxCeleb1	#13/18	In-the-wild	550894	7.4593 0.458766 minDCF
VoxCeleb Short VoxCeleb1	#12/18	Short-duration	394724	22.3595 0.801817 minDCF
3D-Speaker Device 3D-Speaker	#17/18	Channel/device	180000	36.8213 0.999900 minDCF
3D-Speaker Distance 3D-Speaker	#17/18	Distance	175163	37.8673 0.999921 minDCF
3D-Speaker Dialect 3D-Speaker	#17/18	Accent/dialect	180000	38.2767 0.999867 minDCF
FFSVC 2022 Cross-Channel FFSVC 2022	#18/18	Channel/device	72000	26.7500 0.998667 minDCF
FFSVC 2022 Cross-Domain FFSVC 2022	#18/18	Distance	66546	26.9545 0.999067 minDCF
Whisper40 Whisper Whisper40	#18/18	Speaking style	17600	33.6875 0.997938 minDCF
Lombard Grid Lombard Lombard Grid	#17/18	Speaking style	29524	7.9993 0.495715 minDCF
VOiCES Noise/Reverb VOiCES	#16/18	Noise/reverb	55000	20.9800 0.741000 minDCF
CHiME-6 Domestic Far-Field CHiME-6	#17/18	Distance	36487	26.7802 0.970516 minDCF
CHiME-6 Overlap CHiME-6	#16/18	Overlap	39600	29.4111 0.923250 minDCF
ESD ESD	#16/18	Speaking style	437408	12.8825 0.923664 minDCF
AliMeeting Near/Far AliMeeting	#17/18	Distance	220000	23.7800 0.959010 minDCF
AliMeeting Overlap AliMeeting	#18/18	Overlap	165000	23.5933 0.965647 minDCF
VoxKnesset VoxKnesset	#18/18	Aging	158312	27.7733 0.972844 minDCF
VoxPopuli Aging VoxPopuli	#17/18	Aging	146575	10.9824 0.743985 minDCF