Speech recognition using hidden Markov models with polynomial regression functions as nonstationary states

By Li Deng, M. Aksmanovic, Xiaodong Sun, C. F. J. Wu. 1994

How researchers see this paper

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

Citations in the corpus

Deep Learning: Methods and Applications (2014, background): “Specifically, the early work described in [71, 72, 83, 84, 99, 274] generalized and extended the conventional shallow and conditionally independent HMM structure by imposing dynamic constraints, in the form of polynomial trajectory, on the HMM parameters.”
A tutorial survey of architectures, algorithms, and applications for deep learning (2014, methodology): “were a popular approach but had not been competitive with the GMM-HMM (Waibel et al., 1989; Bourlard and Morgan, 1993; Deng et al., 1994; Morgan, 2012).”
Speech trajectory discrimination using the minimum classification error learning (1998, methodology): “The model parameters of the trended HMM, especially the state-dependent time-varying Gaussian means, used in the past were trained by a modified Viterbi algorithm based on the joint-state maximum likelihood (ML) principle [4].”
Deep segmental neural networks for speech recognition (2013, methodology): “It has been known for decades that modeling speech using the conventional HMM has several limitations as analyzed in [9, 10, 11].”
Dynamic Speech Models: Theory, Algorithms, and Applications (2006, methodology): “Importantly, the polynomial trajectories, which were shown to fit well to the temporal properties of cepstral features [55, 56], are not appropriate for the hidden dynamics that require realistic physical constraints of segment-bound monotonicity and targetdirectedness.”
Switching Dynamic System Models for Speech Articulation and Acoustics (2004, background): “Work reported and surveyed in [10, 15, 33, 42] have also focused on the dynamic aspects in the speech process, but the dynamic object being modeled is in the space of speech acoustics, rather than in the space of the production-aÆliated variables.”
An Overview of Deep-Structured Learning for Information Processing (2011, background): “Specifically, the early work described in (Deng 1992, 1993; Deng et al., 1994; Ostendorf et al., 1996, Deng and Sameti, 1996) generalized and extended the conventional shallow HMM structure by imposing dynamic constraints, in the form of polynomial trajectory, on the HMM parameters.”
Structured speech modeling (2006, background): “Importantly, the polynomial trajectories, which were shown to fit well to the temporal properties of cepstral features [18], [35], are not appropriate for the hidden dynamics since they do not generally satisfy at least two physical constraints required for realistic articulation-like dynamics.”
Hidden semi-Markov models (2010, unknown): “q 1 2”
A dynamic, feature-based approach to the interface between phonology and phonetics for speech modeling and recognition (1998, methodology): “The boundaries are estimated using the modi®ed Viterbi algorithm similar to the one described in [23] except that the continuity constraint is imposed across neighboring units via use of nuisance parameter a as discussed above (a is not trained; it is discarded after use for providing the continuity”
From HMM's to segment models: a unified view of stochastic modeling for speech recognition (1996, methodology): “Deng and colleagues extended the parametric trajectory model [15] to the Gauss-Markov case [63], providing the rst positive results for the Gauss-Markov assumption with cepstral derivatives.”