HMM-based speech recognition using state-dependent, discriminatively derived transforms on mel-warped DFT features

By R. Chengalvarayan, Li Deng. 1997

How researchers see this paper

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

1. Li Deng
2. Dong Yu
3. Xiao Li
4. Jinyu Li
5. Xiaodong He
6. Yajie Miao
7. Yoshua Bengio
8. Tara N. Sainath

Citations in the corpus

• Deep Learning: Methods and Applications (2014, methodology): “That is, rather than using Mel-warped filter-bank features as the input features as in [1, 3, 50, 221], the weights corresponding to the Mel-scale filters are only used to initialize the parameters, which are subsequently learned together with the rest of the deep network as the classifier.”
• Large-vocabulary speech recognition under adverse acoustic environments (2000, unknown): “Based on such experience, one promising direction is to extend the discriminative strategy [ 7 ] for joint optimization of the HMM and preprocessor parameters to include also the parameters that characterize noise reduction algorithms.”
• Computational Models for Speech Production (2018, background): “The computation of the gradient is lengthy and laborious and is not included here, but the general spirit of such computation can be gleaned from the work published in [26] applied to a far less sophisticated speech model.”
• Machine Learning Paradigms for Speech Recognition: An Overview (2013, background): “One earliest work in this area was from [125], where dimensionality reduction in the Mel-warped discrete Fourier transform (DFT) feature space was investigated subject to maximal preservation of speech classification information.”
• Discriminative Training for Large-Vocabulary Speech Recognition Using Minimum Classification Error (2007, methodology): “Still within speech recognition, there has been significant work applying MCE to feature extraction [35], [36].”