We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.SD

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Sound

Title: Improved Accent Classification Combining Phonetic Vowels with Acoustic Features

Authors: Zhenhao Ge
Abstract: Researches have shown accent classification can be improved by integrating semantic information into pure acoustic approach. In this work, we combine phonetic knowledge, such as vowels, with enhanced acoustic features to build an improved accent classification system. The classifier is based on Gaussian Mixture Model-Universal Background Model (GMM-UBM), with normalized Perceptual Linear Predictive (PLP) features. The features are further optimized by Principle Component Analysis (PCA) and Hetroscedastic Linear Discriminant Analysis (HLDA). Using 7 major types of accented speech from the Foreign Accented English (FAE) corpus, the system achieves classification accuracy 54% with input test data as short as 20 seconds, which is competitive to the state of the art in this field.
Comments: International Congress on Image and Signal Processing (CISP) 2015
Subjects: Sound (cs.SD); Computation and Language (cs.CL)
DOI: 10.1109/CISP.2015.7408064
Cite as: arXiv:1602.07394 [cs.SD]
  (or arXiv:1602.07394v1 [cs.SD] for this version)

Submission history

From: Zhenhao Ge [view email]
[v1] Wed, 24 Feb 2016 04:33:49 GMT (520kb,D)

Link back to: arXiv, form interface, contact.