Table 6 The best reported phone error rates (PER) on the TIMIT core test set
From: Phone recognition with hierarchical convolutional deep maxout networks
Method | PER (%) |
|---|---|
Hierachical shallow ANN (CD) [25] | 21.2 |
DNN with softmax units [3] | 20.7 |
DNN with ReLU units [35] | 20.8 |
DNN with ReLU units (CD) [35] | 19.8 |
DNN with dropout [28] | 19.7 |
DNN with BMMI features (CD) [50] | 19.1 |
CNN with speaker adaptation [51] | 18.9 |
DNN + RNN [49] | 18.8 |
CNN with heterogeneous pooling [8] | 18.7 |
LSTM RNN [52] | 17.7 |
CNN with scatter features (CD) [53] | 17.4 |
Hierarchical CNN (CD) [this paper] | 16.5 |