Speech Prediction on TIMIT (val)
2.86MSELipschitz RNN
Evaluation Results
| Method | Links | |
|---|---|---|
| Lipschitz RNNNumber of layers=1, Hidden dimension (N)=256, Number of parameters=≈198K, Integration scheme=RK22020.06 | 2.86 | |
| Lipschitz RNNNumber of layers=1, Hidden dimension (N)=256, Number of parameters=≈198K, Integration scheme=Euler2020.06 | 2.95 | |
| Exponential RNN (Lezcano-Casado & Martinez-Rubio, 2019)Number of layers=1, Hidden dimension (N)=425, Number of parameters=≈200K2020.06 | 5.52 | |
| SRLSTM (Nguyen et al., 2020)Number of layers=1, Hidden dimension (N)=158, Number of parameters=≈200K2020.06 | 5.81 | |
| MomentumLSTM (Nguyen et al., 2020)Number of layers=1, Hidden dimension (N)=158, Number of parameters=≈200K2020.06 | 5.86 | |
| Cayley RNN (Helfrich et al., 2018)Number of layers=1, Hidden dimension (N)=425, Number of parameters=≈200K2020.06 | 7.97 | |
| LSTM (Nguyen et al., 2020)Number of layers=1, Hidden dimension (N)=158, Number of parameters=≈200K2020.06 | 9.33 | |
| CConv-LSTM#PARAMS=≈ 88K2017.05 | 10.78 | |
| Conv-LSTM#PARAMS=≈ 88K2017.05 | 11.1 | |
| LSTM (Helfrich et al., 2018)Number of layers=1, Hidden dimension (N)=158, Number of parameters=≈200K2020.06 | 13.66 | |
| Full-capacity Unitary RNN (Wisdom et al., 2016)Number of layers=1, Hidden dimension (N)=256, Number of parameters=≈200K2020.06 | 14.41 | |
| Full-Capacity uRNN#PARAMS=≈ 135K2017.05 | 14.56 | |
| LSTM#PARAMS=≈ 135K2017.05 | 16.59 |