Applying Convolutional Neural Networks concepts to hybrid NN-HMM model for speech recognition

By Ossama Abdel-Hamid, Abdel-rahman Mohamed, Hui Jiang, Gerald Penn. 2012

942 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Model Architectures, Speech & Audio, 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

In Model Architectures

In Speech & Audio

In Signal Processing

Citations in the corpus

Speech recognition with deep recurrent neural networks (2013, background): “Another interesting direction would be to combine frequencydomain convolutional neural networks [27] with deep LSTM.”
Benchmarking Neural Network Robustness to Common Corruptions and Perturbations (2019, background): “Robustness in noisy environments requires robust architectures, and some research finds convolutional networks more robust than fully connected networks (Abdel-Hamid et al., 2013).”
Unsupervised Speech Recognition (2021, background): “Such models are often classified into two categories: hybrid (Juang et al., 1986; Young, 1996; Povey, 2005; Hinton et al., 2012; Abdel-Hamid et al., 2012; Bourlard and Morgan, 2012) or end-to-end (Graves et al., 2006; Graves, 2012; Chorowski et al., 2015; Rao et al., 2017).”
Deep Speech 2 : End-to-End Speech Recognition in English and Mandarin (2015, background): “Recurrent neural networks, typically LSTMs, are just beginning to be deployed in state-of-the art recognizers [24, 25, 55] and work well together with convolutional layers for the feature extraction [52].”
Do Deep Nets Really Need to be Deep? (2013, background): “h 1M labeled points. When you train a shallow neural net with one fully-connected feed-forward hidden layer on this data you obtain 86% accuracy on test data. When you train a deeper neural net as in [1] consisting of a convolutional layer, pooling layer, and three fully-connected feed-forward layer”
Convolutional Neural Network Architectures for Matching Natural Language Sentences (2014, methodology): “As in most convolutional models [11, 1], we use convolution units with a local “receptive ﬁeld” and shared weights, but we design a large feature map to adequately model the rich structures in the composition of words.”
Recent advances in convolutional neural networks (2015, methodology): “In [306, 309, 310], they achieve lower speech recognition errors by applying CNN on Mel filter bank features.”
Residual LSTM: Design of a Deep Recurrent Architecture for Distant Speech Recognition (2017, background): “For example, convolutional neural networks (CNN) which has been huge success in image classification and detection were effective to reduce environmental and speaker variability in acoustic features [6, 7, 8, 9, 10].”
Deep Learning: Methods and Applications (2014, methodology): “It was not until recently that researchers have discovered that the timedimension invariance is less important than the frequency-dimension invariance for speech recognition [1, 3, 81].”
Achieving Human Parity in Conversational Speech Recognition (2016, methodology): “Compared to earlier applications of CNNs to speech recognition [67, 68], our networks are much deeper, and use linear bypass connections across convolutional layers.”
Multimodal Convolutional Neural Networks for Matching Image and Sentence (2015, methodology): “Moreover, as most convolutional models [1, 14, 22], we consider the convolution unit with a local “receptive field” and shared weights to adequately model the rich structures for word composition and inter-modal interaction.”
The microsoft 2016 conversational speech recognition system (2016, methodology): “Compared to earlier applications of CNNs to speech recognition [38, 39], our networks are much deeper, and use linear bypass connections across convolutional layers.”
Estimating phoneme class conditional probabilities from raw speech signal using convolutional neural networks (2013, background): “This aspect needs to be further investigated in the context of large database and using deeper architectures [11], where the slight advantage of MFCC might collapse.”
Analysis of CNN-based speech recognition system using raw speech as input (2015, methodology): “More specifically, use of context-dependent phonemes [4, 5]; use of spectral features as opposed to cepstral features [6, 7]; CNN-based system with Mel filter bank energies as input [8, 9, 10]; combination of different features [11], to name a few.”
Deep Learning for Time-Series Analysis (2017, methodology): “In [1], the features learned by the CNN are used as input to a Hidden Markov Model, achieving a drop at the error rate of over 10%.”
Unsupervised learning of invariant representations (2016, background): “works [3, 4, 31, 32] and Deep Feedforward Neural Networks [33, 34, 35] are examples of this class of architectures—as is, possibly, the feedforward organization of the ventral stream.”
Unsupervised Learning of Invariant Representations in Hierarchical Architectures (2013, background): “Deep learning convolutional networks are another case of architectures that have achieved very good performance and are probably special cases of the theory even if they do not incorporate all of the possible invariances or their unsupervised learning ([44, 45], but see [46]).”