List your company

Universal Speech Representation Evaluation on SUPERB Benchmark

1,010.29Overall ScoreSpeech-FT
4.5164265.6307526.745787.8593Oct 26, 2021Jun 4, 2022Jan 11, 2023Aug 20, 2023Mar 28, 2024Nov 4, 2024Jun 14, 2025

Evaluation results

MethodLinks
Speech-FTFine-tuning Task=ER on IEMOCAP2025.021,010.290.8263—————————87.2385.95——————25.84.95
Speech-FTModel=WavLM Base+, Fine-tuning task=ASR on TED-LIUM2025.02950.60.88124.343.84—5.18—97.440.109—99.190.7369.02——————21.13.42
Pre-trainedModel=WavLM Base+, Fine-tuning task=ASR on TED-LIUM2025.02946.950.89434.123.5—5.57—97.360.0989—98.9990.5268.67——————21.33.94
Speech-FTModel=WavLM Base+, Fine-tuning Task=ASR on TED-LIUM2025.02945.5————————————————————
Pre-trainedModel=WavLM Base+2025.02942.96————————————————————
Stable-FTFine-tuning Task=ER on IEMOCAP2025.02908.860.7307—————————81.490.07——————36.6216.11
Stable-FTModel=WavLM Base+, Fine-tuning Task=ASR on TED-LIUM2025.02907.31————————————————————
Speech-FTFine-tuning Task=ASR on TED-LIUM2025.02905.790.8411—————————88.8467.78——————23.463.94
Stable-FTModel=WavLM Base+, Fine-tuning task=ASR on TED-LIUM2025.02905.090.8325.045.19—6.29—96.720.1189—98.7989.4167.63——————22.953.93
Speech-FTFine-tuning Task=PR on Librispeech2025.02883.930.8195—————————88.665.79——————24.522.99
Speech-FTFine-tuning Task=ASR on Librispeech2025.02882.440.8279—————————87.9766.38——————24.933.93
Speech-FTFine-tuning Task=SID on Voxceleb12025.02881.070.888—————————87.266.03——————25.236.91
Speech-FTFine-tuning Task=SID on Librispeech2025.02877.970.8324—————————87.9865.48——————24.95.67
Speech-FTFine-tuning Task=PC on TIMIT2025.02877.660.8178—————————88.6565.48——————24.054.76
Speech-FTModel=HuBERT, Fine-tuning task=ASR on TED-LIUM2025.02874.510.84115.275.99—5.75—96.690.1011—98.9788.8467.78——————23.463.94
Speech-FTFine-tuning Task=ER on CREMA-D2025.02871.930.8336—————————88.4165.59——————24.755.02
Stable-FTFine-tuning Task=ASR on TED-LIUM2025.02870.710.7989—————————88.5265.38——————23.85.22
Pre-trainedFine-tuning Task=None2025.02870.20.8186—————————88.5464.99——————24.75.17
Speech-FTModel=wav2vec 2.0, Fine-tuning Task=ASR on TED-LIUM2025.02864.79————————————————————
Stable-FTFine-tuning Task=SID on Librispeech2025.02847.330.7552—————————84.7359.51——————31.5414.18
Stable-FTModel=HuBERT, Fine-tuning task=ASR on TED-LIUM2025.028450.79895.346.02—7.63—96.920.097—98.7188.5265.38——————23.85.22
Stable-FTFine-tuning Task=PR on Librispeech2025.02842.030.756—————————88.0963.14——————26.242.21
Pre-trainedModel=wav2vec 2.02025.02837.08————————————————————
Stable-FTModel=wav2vec 2.0, Fine-tuning Task=ASR on TED-LIUM2025.02824.82————————————————————
Mamba + MLPSize=Base, Parameters (M)=94.7, Setting=Causal2025.06823.150.7348——11.72—————89.62—61.72————————
MambaSize=Base, Parameters (M)=78.2, Setting=Causal2025.06817.940.7307——11.68—————87.45—61.98————————
Pre-trainedModel=HuBERT, Fine-tuning task=ASR on TED-LIUM2025.02815.470.81865.566.32—6.38—96.590.0687—98.0288.5464.99——————24.75.17
Causal TransformerSize=Base, Parameters (M)=94.7, Setting=Causal2025.06805.440.6004——13.87—————94.23—63.33————————
Stable-FTFine-tuning Task=ASR on Librispeech2025.02805.330.7214—————————86.0163.82——————29.325.35
Speech-FTModel=DeCoAR 2.0, Fine-tuning Task=ASR on TED-LIUM2025.02781.77————————————————————
Mamba + MLPSize=Small, Parameters (M)=33.0, Setting=Causal2025.06777.680.6842——15.28—————82.39—61.63————————
Causal TransformerSize=Small, Parameters (M)=23.5, Setting=Causal2025.06767.70.5827——15.91—————87.92—62.07————————
MambaSize=Small, Parameters (M)=23.5, Setting=Causal2025.06766.940.6606——17.34—————85.13—60.82————————
Stable-FTFine-tuning Task=ER on CREMA-D2025.02741.190.8148—————————87.2864.69——————27.036.62
Pre-trainedModel=DeCoAR 2.02025.02736.77————————————————————
Transformer + Causal MaskSize=Small, Parameters (M)=23.5, Setting=Causal2025.06735.490.6661——24.05—————81.86—60.45————————
Transformer + Causal MaskSize=Base, Parameters (M)=94.7, Setting=Causal2025.06734.680.7457——25.57—————80.07—58.77————————
Stable-FTFine-tuning Task=PC on TIMIT2025.02726.640.6634—————————83.561.25——————33.8210.34
Stable-FTModel=DeCoAR 2.0, Fine-tuning Task=ASR on TED-LIUM2025.02701.6————————————————————
Stable-FTFine-tuning Task=SID on Voxceleb12025.02651.420.8775—————————77.6460.29——————43.2729.08
WavLM Large#Params=316.62M, Corpus=Mix 94k hr, Weights=Frozen2021.1074.695.493.773.243.063.4432.2797.860.088626.5799.3192.2118.362.794.511.197.3999——
WavLM Base+#Params=94.70M, Corpus=Mix 94k hr, Weights=Frozen2021.1073.489.424.073.53.925.5938.3297.370.098824.259990.5821.22.6394.310.857.48.199——
HuBERT Large#Params=316.61M, Corpus=LL 60k hr, Weights=Frozen2021.1072.290.335.985.753.533.6244.0895.290.035320.0198.7689.8121.762.6494.210.457.22999.25——
WavLM Base#Params=94.70M, Corpus=LS 960 hr, Weights=Frozen2021.107284.514.694.554.846.2142.8196.790.08720.7498.6389.3822.862.589410.377.42898——
WavLM Base (w/o structure modification)#Params=94.68M, Corpus=LS 960 hr, Weights=Frozen2021.1071.984.744.614.725.226.842.8896.790.095620.0398.3188.56242.589410.297.458.499——
WavLM Base (w/o denoising task)#Params=94.70M, Corpus=LS 960 hr, Weights=Frozen2021.1071.784.394.916.034.856.0843.6196.790.079921.0398.4288.6923.432.5693.99.917.437.597.75——
HuBERT Base#Params=94.68M, Corpus=LS 960 hr, Weights=Frozen2021.1070.981.425.115.885.416.4246.6996.30.073615.5398.3488.5325.22.5893.99.367.47898.5——
wav2vec 2.0 Large#Params=317.38M, Corpus=LL 60k hr, Weights=Frozen2021.1070.486.145.655.624.753.7544.6996.660.048912.4895.2887.1127.312.529410.027.6315.897.25——
wav2vec 2.0 Base#Params=95.04M, Corpus=LS 960 hr, Weights=Frozen2021.1069.675.186.026.085.746.4346.9596.230.023314.8192.3588.324.772.5593.99.777.510.598——
DeCoAR 2.0#Params=89.84M, Corpus=LS 960 hr, Weights=Frozen2021.1066.374.427.166.5914.9313.0253.6294.480.04069.9490.883.2834.732.4793.28.547.8317.190.75——
wav2vec#Params=32.54M, Corpus=LS 960 hr, Weights=Frozen2021.1063.556.567.999.931.5815.8655.8695.590.04856.6184.9276.3743.712.5393.89.37.4510.198.25——
vq-wav2vec#Params=34.15M, Corpus=LS 960 hr, Weights=Frozen2021.1061.838.810.389.9333.4817.7160.6693.380.0415.6685.6877.6841.542.4893.68.167.0813.4100——
VQ-APC#Params=4.63M, Corpus=LS 360 hr, Weights=Frozen2021.1059.560.158.7210.4541.0821.263.5691.110.02514.2374.4868.5352.912.5693.48.447.8422.494.25——
APC#Params=4.11M, Corpus=LS 360 hr, Weights=Frozen2021.1059.260.428.5610.5341.9821.2863.1291.010.0315.9574.6970.4650.892.5693.48.928.0527.287.25——
NPC#Params=19.38M, Corpus=LS 360 hr, Weights=Frozen2021.105955.929.49.3443.8120.261.6688.960.02464.3269.4472.7948.442.5293.18.047.8630.494.75——
TERA#Params=21.33M, Corpus=LS 960 hr, Weights=Frozen2021.1057.257.5715.899.9649.1718.1758.4989.480.00135.6658.4267.554.172.5493.610.198.2125.183.75——
modified CPC#Params=1.84M, Corpus=LL 60k hr, Weights=Frozen2021.1056.939.6312.8610.3842.5420.1862.5491.880.03264.8264.0971.1949.912.5793.710.48.4126.271——
PASE+#Params=7.83M, Corpus=LS 50 hr, Weights=Frozen2021.1051.537.9911.618.6858.8725.1161.5682.540.00723.1629.8262.1460.172.5693.99.878.6630.663.2——
Mockingjay#Params=85.12M, Corpus=LS 360 hr, Weights=Frozen2021.105132.2911.6610.5470.1922.8265.2783.670.00074.4534.3361.5958.892.5393.49.298.2935.179.75——
FBANK#Params=0, Corpus=-, Weights=Frozen2021.1043.20.00099.5610.0582.0123.1863.588.630.00582.329.169.6452.942.5593.69.238.4738.377.25——
HubertSize=Base2026.01—0.81425.115.885.416.42—96.30.0736—98.3488.5364.92——————25.2—
Mimi2026.01—0.342312.1713.3739.6988.65—93.120.0108—90.616.7256.05——————98.86—
Spectral (FBANK)2026.01—09.5610.0582.0123.18—8.630.0058—9.169.6435.39——————52.94—
SylberNotes=English-only original version2026.01—0.51248.7515.5588.798.88—97.110.0591—99.0885.6665.25——————29.49—
Sylber 2.0Notes=Universal version2026.01—0.72035.696.114.7410.04—96.850.0471—98.0585.9462.37——————30.09—
Wav2Vec 2.0Size=Base2026.01—0.75186.026.085.746.43—96.230.0233—92.3588.363.43——————24.77—
WavLMSize=Large2026.01—0.95493.773.243.063.44—97.860.0886—99.3192.2170.62——————18.36—