A scalable approach to using DNN-derived features in GMM-HMM based acoustic modeling for LVCSR

By Zhijie Yan, Qiang Huo, Jian Xu. 2013

56 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

In Signal Processing

Citations in the corpus

Deep Learning: Methods and Applications (2014, methodology): “The collaboration continued to large vocabulary tasks with more convincing, highly positive results [67, 68, 94, 89, 161, 199, 195, 223, 323, 353, 399, 414].”
A tutorial survey of architectures, algorithms, and applications for deep learning (2014, background): “, 2013) and large vocabulary speech recognition tasks (Yu et al., 2012; Seide et al., 2011; Dahl et al., 2011, 2012; Kubo et al., 2012; Deng et al, 2013b, 2013c; Yan et al., 2013) by integrating the powerful discriminative training ability of DNNs with the sequential modeling ability of HMMs.”
Hierarchical Bayesian combination of plug-in maximum a posteriori decoders in deep neural networks-based speech recognition and speaker adaptation (2017, methodology): “In Yan et al. (2013), it was argued that there is no longer need for a bottleneck layer, and the output of the last hidden layer can be directly utilised as a DNN-based speech features vector.”
Design and Learning of Output Representations for Speech Recognition (2013, background): “27 In recent years, learning representations using deep models, notably those based on deep neural 28 networks (DNNs), have largely focused on the sensory input data, such as speech and image 29 [20][5][22] with visible successes most notably in industry-scale, real-world speech and image 30 recogni”
Sequence classification using the high-level features extracted from deep neural networks (2014, methodology): “[35] experimentally compare the GMM-HMM and DNN-HMM approaches, using the tandem or top-hidden-layer features.”
Deep learning of feature representation with multiple instance learning for medical image analysis (2014, methodology): “The main methods of feature extraction are manually designed feature descriptors [7, 8], fully supervised feature learning [9] and unsupervised feature learning [10].”
Attention-based Wav2Text with feature transfer learning (2017, methodology): “This idea first trains a DNN with a narrow hidden bottleneck layer to perform phoneme classification at the frame level and then reuses the activations from the narrow hidden bottleneck layer as discriminative features in conventional GMM-HMM or hybrid DNN-HMM models [30].”