Keyword Spotting on Google Speech Commands v1 (test)
98AccuracyBC-ResNet-8
Evaluation Results
| Method | Links | |
|---|---|---|
| BC-ResNet-8#Param=321k, #Mult=89.1M2021.06 | 98 | |
| BC-ResNet-6#Param=188k, #Mult=53.1M2021.06 | 97.9 | |
| BC-ResNet-3#Param=54.2k, #Mult=16.2M2021.06 | 97.6 | |
| MatchboxNet-3x2x64#Param=93k, #Mult=11.3M2021.06 | 97.5 | |
| BC-ResNet-2#Param=27.3k, #Mult=8.5M2021.06 | 97.3 | |
| MHAtt-RNNimplementation=Optimized (SpecAugment + tuning)2020.05 | 97.2 | |
| BC-ResNet-1.5#Param=17.2k, #Mult=5.5M2021.06 | 97.2 | |
| MatchboxNet-3x1x64#Param=77k, #Mult=9.3M2021.06 | 97.2 | |
| MHAtt-RNN#Param=743k, #Mult=22.7M2021.06 | 97.2 | |
| TC-ResNetimplementation=Optimized (SpecAugment + tuning)2020.05 | 97.1 | |
| DSCNN+strdimplementation=Optimized (SpecAugment + tuning)2020.05 | 97 | |
| CRNNimplementation=Optimized (SpecAugment + tuning)2020.05 | 97 | |
| Attention RNNTrainable Parameters=202K, Model Variant=V22018.08 | 96.9 | |
| LSTMimplementation=Optimized (SpecAugment + tuning)2020.05 | 96.9 | |
| EdgeSpeechNet-A# Parameters, K=1072020.04 | 96.8 | |
| DenseNet-BC-100# Parameters, K=8002020.04 | 96.77 | |
| DS-ResNet18#Param=72 k, #Mult=285 M2021.06 | 96.7 | |
| GRUimplementation=Optimized (SpecAugment + tuning)2020.05 | 96.6 | |
| TC-ResNetimplementation=Baseline Reference2020.05 | 96.6 | |
| TC-ResNet14-1.5#Param=305k, #Mult=6.7M2021.06 | 96.6 | |
| TENet12#Param=100k, #Mult=2.9M2021.06 | 96.6 | |
| BC-ResNet-1#Param=9.2k, #Mult=3.1M2021.06 | 96.6 | |
| SVDFimplementation=Optimized (SpecAugment + tuning)2020.05 | 96.3 | |
| S4D-LinParameters=306K, Sampling Rate=16000Hz, Zero-shot=false2022.06 | 96.25 | |
| SSAST 250Masked Patches=250, Pretraining=Self-Supervised2021.10 | 96.2 | |
| S4D-InvParameters=306K, Sampling Rate=16000Hz, Zero-shot=false2022.06 | 96.18 | |
| S4-LegSParameters=307K, Sampling Rate=16000Hz, Zero-shot=false2022.06 | 96.08 | |
| SSAST 400Masked Patches=400, Pretraining=Self-Supervised2021.10 | 96 | |
| DS-ResNet14#Param=15.2k, #Mult=15.7M2021.06 | 95.9 | |
| S4D-LegSParameters=306K, Sampling Rate=16000Hz, Zero-shot=false2022.06 | 95.83 | |
| res15Trainable Parameters=238K2018.08 | 95.8 | |
| ResNet-15#Param=238k, #Mult=894M2021.06 | 95.8 | |
| Attention RNNTrainable Parameters=202K2018.08 | 95.6 | |
| Att-RNNimplementation=Baseline Reference2020.05 | 95.6 | |
| Att-RNN#Param=202k, #Mult=22.3M2021.06 | 95.6 | |
| ConvNetParameters=26.2M, Sampling Rate=16000Hz, Zero-shot=false2022.06 | 95.51 | |
| AST-IM + KDPretraining=ImageNet + Knowledge Distillation2021.10 | 95.5 | |
| DS-CNNTrainable Parameters=498K2018.08 | 95.4 | |
| CNN+strdimplementation=Optimized (SpecAugment + tuning)2020.05 | 95.4 | |
| DSCNN+strdimplementation=Baseline Reference2020.05 | 95.4 | |
| S4-(LegS/FouT)Parameters=307K, Sampling Rate=16000Hz, Zero-shot=false2022.06 | 95.32 | |
| S4-FouTParameters=307K, Sampling Rate=16000Hz, Zero-shot=false2022.06 | 95.27 | |
| res26Trainable Parameters=438K2018.08 | 95.2 | |
| CRNNimplementation=Baseline Reference2020.05 | 95 | |
| LSTMimplementation=Baseline Reference2020.05 | 94.8 | |
| GRUimplementation=Baseline Reference2020.05 | 94.7 | |
| res8Trainable Parameters=110K2018.08 | 94.1 | |
| CNN+strdimplementation=Baseline Reference2020.05 | 92.7 | |
| S4D-InvParameters=306K, Sampling Rate=8000Hz, Zero-shot=true2022.06 | 91.8 | |
| AST-AudioSetPretraining=Supervised AudioSet2021.10 | 91.6 | |
| S4-FouTParameters=307K, Sampling Rate=8000Hz, Zero-shot=true2022.06 | 91.59 | |
| S4D-LinParameters=306K, Sampling Rate=8000Hz, Zero-shot=true2022.06 | 91.58 | |
| S4-LegSParameters=307K, Sampling Rate=8000Hz, Zero-shot=true2022.06 | 91.32 | |
| DNNimplementation=Optimized (SpecAugment + tuning)2020.05 | 91.2 | |
| S4D-LegSParameters=306K, Sampling Rate=8000Hz, Zero-shot=true2022.06 | 91.08 | |
| S4-(LegS/FouT)Parameters=307K, Sampling Rate=8000Hz, Zero-shot=true2022.06 | 90.72 | |
| ConvNet on raw WAVTrainable Parameters=700K2018.08 | 89.4 | |
| AST-ScratchTraining=From scratch2021.10 | 87.2 | |
| DNNimplementation=Baseline Reference2020.05 | 86.7 | |
| XResNet-50Parameters=904K, Sampling Rate=16000Hz, Zero-shot=false2022.06 | 83.01 | |
| ResNet-18Parameters=216K, Sampling Rate=16000Hz, Zero-shot=false2022.06 | 77.86 | |
| TRILLEvaluation Protocol=Linear Classifier2020.10 | 74 | |
| COLAEvaluation Protocol=Linear Classifier2020.10 | 71.7 | |
| InceptionNetParameters=481K, Sampling Rate=16000Hz, Zero-shot=false2022.06 | 61.24 | |
| ResNet-18Parameters=216K, Sampling Rate=8000Hz, Zero-shot=true2022.06 | 8.74 | |
| XResNet-50Parameters=904K, Sampling Rate=8000Hz, Zero-shot=true2022.06 | 7.72 | |
| ConvNetParameters=26.2M, Sampling Rate=8000Hz, Zero-shot=true2022.06 | 7.26 | |
| InceptionNetParameters=481K, Sampling Rate=8000Hz, Zero-shot=true2022.06 | 5.18 |