Comparing performance of acoustic-to-articulatory inversion for mandarin accented english and american english speakers

This paper compares the performance of acoustic-to-articulatory inversion for both L1 and L2 speakers of English, as a function of the number of Gaussian Mixtures used in the inversion model. The inversion system is based on an HMM-GMM approach and is implemented on the Marquette Electromagnetic Art...

Full description

Saved in:
Bibliographic Details
Published in2018 IEEE International Symposium on Signal Processing and Information Technology (ISSPIT) pp. 1 - 5
Main Authors Bozorg, Narjes, Johnson, Michael T.
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.12.2018
Subjects
Acoustic-to-Articulatory Inversion
Acoustics
Articulatory Features
Electro-Magnetic Articulography
Gaussian Mixture Model
Hidden Markov Model
Hidden Markov models
Kinematics
Lips
Sensor phenomena and characterization
Tongue
Online AccessGet full text

Cover

More Information
Summary:This paper compares the performance of acoustic-to-articulatory inversion for both L1 and L2 speakers of English, as a function of the number of Gaussian Mixtures used in the inversion model. The inversion system is based on an HMM-GMM approach and is implemented on the Marquette Electromagnetic Articulography corpus of Mandarin Accented English (EMAMAE) including 20 native English speakers and 19 native Mandarin speakers of English. Results indicate that for Mandarin speakers 12 Gaussian mixtures and for L1 American English speakers 11 Gaussian mixtures give the lowest Root-Mean-Squared error (RMSE) and highest correlation between the estimated and actual articulatory pattern.
DOI:10.1109/ISSPIT.2018.8705093