Use of Differential Cepstra as Acoustic Features in Hidden Trajectory Modeling for Phonetic Recognition

By Li Deng, Dong Yu. 2007

54 citations recorded by the source corpus.

The earlier version of the hidden trajectory model (HTM) for speech dynamics which predicts the "static" cepstra as the observed acoustic feature is generalized to one which predicts joint static cepstra and their temporal differentials (i.e., delta cepstra). The formulation of this generalized HTM is presented in the generative-modeling framework, enabling efficient computation of the joint likelihood for both static and delta cepstral sequences as the acoustic features given the model. The parameter estimation techniques for the new model are developed and presented, giving closed-form estimation formulas after the use of vector Taylor series approximation. We show principled generalization from the earlier static-cepstra HTM to the new static/delta-cepstra HTM not only in terms of model formulations but also in terms of their respective analytical forms in (monophone) parameter estimation. Experimental results on the standard TIMIT phonetic recognition task demonstrate recognition accuracy improvement over the earlier best HTM system, both significantly better than state-of-the-art triphone HMM systems.

Subfields: Artificial Intelligence, Signal Processing

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

Deep Learning: Methods and Applications (2014, methodology): “Further, the DNN results were found to be slightly superior to the then-best-performing single system based on the generative hidden trajectory model (HTM) in the literature [105, 108] evaluated on the same, commonly used TIMIT task by many speech researchers [107, 108, 274, 313].”
Learning a better representation of speech soundwaves using restricted boltzmann machines (2011, methodology): “Large-Margin GMM[14] 30.1% CD-HMM[15] 27.3% Augmented Conditional Random Fields[15] 26.6% Recurrent Neural Nets[16] 26.1% Bayesian Triphone HMM[17] 25.6% Monophone HTMs[ 18 ] 24.8% Heterogenous Classifiers[19] 24.4% Deep Belief Networks (bounding silences ignored)[9] 23.0% DBN using RBM on Raw Speec”
An Overview of Deep-Structured Learning for Information Processing (2011, methodology): “More efficient implementation of this deep architecture with hidden dynamics is achieved with non-recursive or FIR filters in more recent studies (Deng et. al., 2006a. 2006b, Deng and Yu, 2007).”
New types of deep neural network learning for speech recognition and related applications: an overview (2013, background): “[14][15], in the literature and evaluated on the identical task.”
A deep convolutional neural network using heterogeneous pooling for trading acoustic invariance with phonetic confusion (2013, background): “An example of such speech variability is formant undershooting (or overshooting) caused by casual (or forced) speaking styles that have been mathematically represented by hidden dynamic or trajectory models of speech [11][12][13] [14].”
Exemplar-Based Sparse Representation Features: From TIMIT to LVCSR (2011, methodology): “In addition, the IBM CD system also offers improvements over the Hidden Trajectory Model (HTM) [3], which computes the total acoustic model score by combining scores from separate HMM and HTM systems.”