Maximum a posteriori estimation for multivariate Gaussian mixture observations of Markov chains

By Jean-Luc Gauvain, Chin-Hui Lee. 1994

Subfields: Probabilistic & Bayesian Methods

How researchers see this paper

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

In Probabilistic & Bayesian Methods

Citations in the corpus

TED-LIUM 3: twice as much data and corpus repartition for experiments on speaker adaptation (2018, methodology): “The SAT AM was trained using GMM-derived (GMMD) features [14], adapted with maximum a posteriori adaptation (MAP) algorithm [2].”
A Theory of Multiple-Source Adaptation with Limited Target Labeled Data (2020, background): “…of the domain adaptation problem that arises in a variety of di↵erent applications, such as in natural language processing (Blitzer et al., 2007; Dredze et al., 2007; Jiang and Zhai, 2007), speech processing (Gauvain and Lee, 1994; Jelinek, 1997), and computer vision (Leggetter and Woodland, 1995).”
Agnostic Federated Learning (2019, background): “, 2018), speech recognition (Legetter and Woodland, 1995; Gauvain and Chin-Hui, 1994; Pietra et al., 1992; Rosenfeld, 1996; Jelinek, 1998; Roark and Bacchiani, 2003), and computer vision (Martı́nez, 2002) A problem more closely related to that of federated learning is that of multiple-source adaptat”
Multiple Source Adaptation and the Rényi Divergence (2009, background): “m a target domain. This problem arises in a variety of natural language processing tasks such parsing, statistical language modeling, text classiﬁcation [15, 7, 16, 9, 10, 4, 6], or speech processing [11, 8, 14] and computer vision [13] tasks, as well as in many other applications. Several recent st”
Probabilistic Elastic Matching for Pose Variant Face Verification (2013, methodology): “also use a Normal-Wishart distribution over the variance as in [9, 10].”
Rich Prosody Diversity Modelling with Phone-Level Mixture Density Network (2021, methodology): “In traditional ASR systems, one of the most dominant techniques is HMM-GMM [15, 16, 17], in which the distribution of acoustic features for each HMM state is modeled with a GMM.”
Large Vocabulary Continuous Speech Recognition : a ReviewSteve (1996, methodology): “Finally, when there is a reasonable amount of adaptation data classical MAP estimation can be used to update all of the acoustic parameters[64, 37].”
Data augmentation for low resource languages (2014, methodology): “The MAP estimation introduces a prior on model parameters into training objective function [6, 13].”
Phone-Level Prosody Modelling With GMM-Based MDN for Diverse and Controllable Speech Synthesis (2021, methodology): “In traditional automatic speech recognition (ASR) systems, one of the most dominant techniques is HMM-GMM [24], [25], [26], in which the distribution of acoustic features for each HMM state is modeled with a GMM. Similarly, GMM is also used to model acoustic features in traditional SPSS [27], [28]…”
Domain Adaptation with Multiple Sources (2008, background): “This set-up and the type of solutions just described have been in fact explored extensively in applications [8, 9, 11, 13–15].”
Analysis of Speaker Adaptation Algorithms for HMM-Based Speech Synthesis and a Constrained SMAPLR Adaptation Algorithm (2009, background): “This is consistent with the nature of MAP estimation [24].”
Supervised and unsupervised PCFG adaptation to novel domains (2003, methodology): “We follow Gauvain and Lee (1994) in adopting a parameter tying approach.”
Mixture Density Network for Phone-Level Prosody Modelling in Speech Synthesis (2021, methodology): “In traditional ASR systems, one of the most dominant techniques is HMM-GMM [14, 15, 16], in which the distribution of acoustic features for each HMM state is modeled with a GMM. Similarly, GMM is also used to model acoustic features in traditional statistical parametric speech synthesis(SPSS) [17,…”
Power efficient processor architecture and the cell processor (2005, background): “Figure 1 shows the organization of the first-generation Cell processor [5].”
Model-Based Expectation-Maximization Source Separation and Localization (2010, methodology): “Because the ILD is modeled with a Gaussian, we use the normal-Wishart distribution, its conjugate, as the prior [25].”