Investigation of deep neural networks (DNN) for large vocabulary continuous speech recognition: Why DNN surpasses GMMS in acoustic modeling

By Jia Pan, Cong Liu, Zhiguo Wang, Yu Hu, Hui Jiang. 2012

103 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Dong Yu
2. Li Deng
3. Hui Jiang
4. Jinyu Li
5. Ossama Abdel-Hamid
6. Abdel-rahman Mohamed
7. Yoshua Bengio
8. Jianfeng Gao

In Signal Processing

1. Dong Yu
2. Li Deng
3. Jinyu Li
4. Ossama Abdel-Hamid
5. Hui Jiang
6. Abdel-rahman Mohamed
7. Lirong Dai
8. Tara N. Sainath

Citations in the corpus

• Leverage Financial News to Predict Stock Price Movements Using Word Embeddings and Deep Neural Networks (2015, background): “For examples, [2, 37, 39, 40] show significant improvements on the worderror rate of speech-to-text transcription task, which was contributed by the effective and”
• Fast speaker adaptation of hybrid NN/HMM model for speech recognition based on discriminative learning of speaker code (2013, methodology): “Recently, the hybrid NN-HMM model has gained more research interests in speech recognition because it has achieved very good performance in both small tasks like TIMIT [8] and large vocabulary tasks [9, 10, 11, 12] especially when a deep NN (DNN) is used.”
• Rectified linear neural networks with tied-scalar regularization for LVCSR (2015, methodology): “We follow the same training procedure as described in [17, 18, 19, 20] to train the conventional context dependent DNN/HMM with the tied-triphone state alignment obtained from the above MLE trained GMM/HMMs baseline system.”
• Convolutional Neural Networks for Speech Recognition (2014, background): “O. Abdel-Hamid and H. Jiang are with the Department of Electrical Engi­neering and Computer Science, Lassonde School of Engineering, York Univer­sity, Toronto, ON M3J 1P3, Canada (e-mail: ossama@cse.yorku.ca; hj@cse. yorku.ca).”
• Deep neural network with weight sparsity control and pre-training extracts hierarchical features and enhances classification performance: Evidence from whole-brain resting-state functional connectivity patterns of schizophrenia (2015, unknown): “Empirical evidence suggests that the DNN can efficiently handle correlated features (Pan et al., 2012), although a rigorous theoretical derivation examining how it does so is warranted.”