The subspace Gaussian mixture model - A structured model for speech recognition

By Daniel Povey, Lukáš Burget, Mohit Agarwal, Pinar Akyazi, Kai Feng, Arnab Ghoshal, Ondrej Glembek, Nagendra Kumar Goel, Martin Karafiát, A. Rastrow and 3 more. Computer Speech and Language, 2011

323 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Signal Processing

View Paper: https://www.semanticscholar.org/paper/8d73b5b6bbb93fdfb3974632b72930741cdedf46

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

Context-dependent modelling of deep neural network using logistic regression (2013, background): “The proposed model is similar to Canonical State Modelling (CSM) [13] for CD GMM/HMM systems and Subspace 338978-1-4799-2756-2/13/$31.00 ©2013 IEEE ASRU 2013 GMM (SGMM) [17], where the CD states are the transformed versions of one or more canonical states.”
The Kaldi Speech Recognition Toolkit (2011, unknown): “in [13].”
Full-covariance UBM and heavy-tailed PLDA in i-vector speaker verification (2011, unknown): “Σ̂ (c) ← floor(Σ, fΣ) implemented using floor function defined in [5]:”
Multilingual training of deep neural networks (2013, methodology): “The subspace Gaussian mixture model (SGMM) [8] has been shown to be suitable for cross-lingual modeling without requiring an explicit mapping between phone units in different languages [9, 10].”
A pitch extraction algorithm tuned for automatic speech recognition (2014, methodology): “Our current recipe is based on combination of three types of Kaldi system, but for speed we only test one of them here: it is a Subspace Gaussian Mixture Model (SGMM) system [12], discriminatively trained with boosted MMI (bMMI) [13].”
Improving deep neural network acoustic models using generalized maxout networks (2014, methodology): “We also compare with the baseline SGMM+bMMI system.”
Context dependent recurrent neural network language model (2012, result): “Note that while we do not use advanced acoustic modeling techniques such as SGMMs [36] and speaker adaptation, the baseline system achieves comparable results as is common in the language modeling research [37] and is sufficient for comparison of advanced language modeling techniques.”
Joint Uncertainty Decoding for Noise Robust Subspace Gaussian Mixture Models (2013, methodology): “Recent research has indicated that an SGMM acoustic model may result in 1558-7916 © 2013 IEEE more accurate speech recognition compared with its GMM counterpart, in both monolingual and multilingual settings [17]–[21].”
Deep maxout networks for low-resource speech recognition (2013, methodology): “On top of the SAT systems, we train subspace Gaussian mixture models (SGMM) [18] for better recognition outcomes.”
iVector-based discriminative adaptation for automatic speech recognition (2011, methodology): “To define the regions in the acoustic space, all Gaussians from ML trained HMM model are pooled and clustered using agglomerative clustering to create GMM with desired number of components (see [19] for detailed description of the clustering algorithm).”
Constructing long short-term memory based deep recurrent neural networks for large vocabulary speech recognition (2014, methodology): “GMM 48.68 Subspace GMM 44.29 ReLU DNN 38.42 PNorm DNN 38.01 Experiments were conducted to evaluate these deep LSTM networks shown in Fig.”
Articulatory Control of HMM-Based Parametric Speech Synthesis Using Feature-Space-Switched Multiple Regression (2013, unknown): “A similar model structure can be found in subspace GMMmodelling [28], where all HMM states share the same GMM structure and the state-dependent subspace vectors play the role of external articulatory features in our method.”