MDL-based context-dependent subword modeling for speech recognition

By K. Shinoda, Takao Watanabe. 2000

302 citations recorded by the source corpus.

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

Deep Clustered Convolutional Kernels (2015, methodology): “We want to address this basic scalability issue of the deep learning development pipeline with training methods that automatically search for DNN architectures while jointly estimating model parameters.”
Speech Synthesis Based on Hidden Markov Models (2013, methodology): “The state tying process is conducted in a hierarchical tree structure manner, and the tree size is automatically determined based on an information criterion called minimum description length (MDL) [57].”
Statistical Parametric Speech Synthesis (2007, methodology): “Minimum description length (MDL) (Rissanen, 1980) criterion-based phonetic decision-tree clustering (Shinoda and Watanabe, 2000) has been used in the HMM-based speech synthesis system to balance model complexity and accuracy.”
Robust Speaker-Adaptive HMM-Based Text-to-Speech Synthesis (2009, methodology): “Then, using the state occupancy probabilities obtained in the SAT framework, decision-tree-based context clustering (using a minimum description length (MDL) criterion [55]) is applied to the HSMMs,”
Unsupervised learning for text-to-speech synthesis (2013, methodology): “For these systems, the Minimum Description Length approach described in Shinoda and Watanabe (2000) is used to determine when tree-building should cease.”
Modeling Spectral Envelopes Using Restricted Boltzmann Machines and Deep Belief Networks for Statistical Parametric Speech Synthesis (2013, methodology): “…problem of the context-dependent model training with extensive context features , a decision-tree-based model clustering technique that uses a minimum description length (MDL) criterion [36] to guide the tree construction is adopted after initial training of the context-dependent HMMs.”
An HMM-based speech synthesis system applied to English (2002, methodology): “To overcome this problem, a decision-tree based context clustering technique [11, 12] is applied to distributions for spectrum, F0 and state duration in the same manner as HMM-based speech recognition.”
Autoregressive Models for Statistical Parametric Speech Synthesis (2013, methodology): “We refer to the combination of modeling with an HMM during training but restricting to during synthesis as the standard HMM synthesis framework.”
Statistical approach to vocal tract transfer function estimation based on factor analyzed trajectory HMM (2008, methodology): “And then, we constructed tied-state triphone HMMs using a decision-tree based context clustering technique adopting the minimum description length (MDL) criterion [8].”
Recent development of the HMM-based singing voice synthesis system - Sinsy (2010, methodology): “The MDL criterion [20] was used to control the size of the decision trees.”
TTS synthesis with bidirectional LSTM based recurrent neural networks (2014, methodology): “Minimum description length (MDL) criterion [17] for balancing model complexity and training data size is used as a stopping criterion for state clustering in decision tree growing.”
Analysis of Speaker Adaptation Algorithms for HMM-Based Speech Synthesis and a Constrained SMAPLR Adaptation Algorithm (2009, methodology): “Then, shared-decision-tree-based context clustering [19] (using a minimum description length (MDL) criterion [64]) was applied to the HSMMs, and the model parameters of the HSMMs at each leaf node of the decision trees were tied.”
Articulatory Control of HMM-Based Parametric Speech Synthesis Using Feature-Space-Switched Multiple Regression (2013, unknown): “In the and systems, the optimal numbers of GMM mixture components for the feature-space regression matrix switching were determined using the minimum description length criterion [27].”
Postfilters to Modify the Modulation Spectrum for Statistical Parametric Speech Synthesis (2016, methodology): “…functions over multiple frames with the HMM-state-level probability density function, which is usually determined by decision tree clustering based on the Minimum Description Length (MDL) criterion [49], CLUSTERGEN predicts the probability density functions frame by frame in the CART framework.”