The Application of Hidden Markov Models in Speech Recognition

By Mark John Francis Gales, Susannah Young. Foundations and Trends® in Signal Processing, 2007

910 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/9eab007a8c0af72f1fdffe2aef210d790fb79d79

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

Citations in the corpus

Brain-to-text: decoding spoken phrases from phone representations in the brain (2015, background): “Simpliﬁed, the process can be understood (Gales and Young, 2008) as ﬁnding the sequence of words W = w 1 . . . w L which is most likely given the observed ECoG feature segments X = F 1 . . .”
Efficient Implementation of Recurrent Neural Network Transducer in Tensorflow (2018, result): “In contrast to the conventional approach [1, 2] where these models are trained separately, all-neural approaches promise a simplified end-to-end speech recognition model.”
Current trends in multilingual speech processing (2011, background): “…speech synthesis (Zen et al 2009) has resulted in a flurry of new research activities and is contributing to substantial efforts to ‘cross-pollinate’ ideas between ASR and TTS, as techniques in these two fields have become increasingly interrelated (Ostendorf & Bulyko 2002; Gales & Young 2007).”
Word-level Acoustic Modeling with Convolutional Vector Regression (2012, background): “This basic system has been refined over the past 25 years, resulting in extremely complicated training recipes to achieve state-of-the-art recognition performance (Gales & Young, 2007).”
Multitask Learning with Low-Level Auxiliary Tasks for Encoder-Decoder Based Speech Recognition (2017, background): “For example, traditional ASR systems include components like frame classifiers, phonetic acoustic models, lexicons (which may or may not be learned from data), and language models [1].”
NAUTILUS: A Versatile Voice Cloning System (2020, background): “dialog systems, are placed at the two ends of the human-machine communication interface and compliment each other, historically, they are built independently under different frameworks [1], [51].”
Theory and Use of the EM Algorithm (2011, methodology): “5 More on HMM and EM For further details on HMM, we recommend Rabiner’s tutorial [49] and the review article by Gales and Young [16], which considers the practical application of HMMs in depth.”
Building DNN acoustic models for large vocabulary speech recognition (2014, methodology): “A full overview of LVCSR systems is beyond the scope of this work, so we instead refer to previous articles for an overview of HMM-based speech recognition systems [8], [9], [10], [11], [12].”
Multiple Change-Point Estimation With a Total Variation Penalty (2010, methodology): “Yet approximate dynamic programming procedures were devised in other contexts, such as for Dynamic Time Warping or the Viterbi algorithm (Kolesnikov and Fraenti 2003; Gales and Young 2008).”
Efficient Orthogonal Parametrisation of Recurrent Neural Networks Using Householder Reflections (2016, background): “Some of these were ”toy” datasets specifically designed (Gers et al. (2001)) to test RNNs for their ability to learn long-term dependencies.”
Speech Synthesis Based on Hidden Markov Models (2013, methodology): “Since the amount of adaptation data is limited, a regression class tree is normally used to cluster the Gaussian components based on acoustic similarity and to share the same MLLR transform [62].”