Vocal Tract Length Invariant Features for Automatic Speech Recognition

Alfred Mertins; Jan Rademacher

doi:10.1109/ASRU.2005.1566473

Vocal Tract Length Invariant Features for Automatic Speech Recognition

Alfred Mertins, Jan Rademacher

Institute of Signal Processing

University of Oldenburg

Abstract

The effects of vocal tract length (VTL) variation are often approximated by linear frequency warping of short-time spectra. Based on this relationship, we present a method for generating vocal tract length invariant features. These new features are computed as translation invariant, correlation-type features in a log-frequency domain. In phoneme classification experiments, their discrimination capabilities turned out to be considerably better than for Mel-frequency cepstral coefficients (MFCCs). The best results are obtained when VTL-invariant (VTLI) features and MFCCs are combined. The superiority of the combined feature set and its resilience to VTL variations is also shown for word recognition, using the TIDIGITS corpus and the HTK recognizer.

Original language	English
Pages	33-37
Number of pages	5
DOIs	https://doi.org/10.1109/ASRU.2005.1566473
Publication status	Published - 01.12.2005
Event	2005 IEEE Automatic Speech Recognition and Understanding Workshop - Cancun, Mexico Duration: 27.11.2005 → 01.12.2005 Conference number: 68918

Conference

Conference	2005 IEEE Automatic Speech Recognition and Understanding Workshop
Abbreviated title	ASRU 2005
Country/Territory	Mexico
City	Cancun
Period	27.11.05 → 01.12.05

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.1109/ASRU.2005.1566473

Cite this

@conference{c0b8f58fb0d440b6bc34c9bff0b85e96,

title = "Vocal Tract Length Invariant Features for Automatic Speech Recognition",

abstract = "The effects of vocal tract length (VTL) variation are often approximated by linear frequency warping of short-time spectra. Based on this relationship, we present a method for generating vocal tract length invariant features. These new features are computed as translation invariant, correlation-type features in a log-frequency domain. In phoneme classification experiments, their discrimination capabilities turned out to be considerably better than for Mel-frequency cepstral coefficients (MFCCs). The best results are obtained when VTL-invariant (VTLI) features and MFCCs are combined. The superiority of the combined feature set and its resilience to VTL variations is also shown for word recognition, using the TIDIGITS corpus and the HTK recognizer.",

author = "Alfred Mertins and Jan Rademacher",

year = "2005",

month = dec,

day = "1",

doi = "10.1109/ASRU.2005.1566473",

language = "English",

pages = "33--37",

note = "2005 IEEE Automatic Speech Recognition and Understanding Workshop ; Conference date: 27-11-2005 Through 01-12-2005",

}

TY - CONF

T1 - Vocal Tract Length Invariant Features for Automatic Speech Recognition

AU - Mertins, Alfred

AU - Rademacher, Jan

N1 - Conference code: 68918

PY - 2005/12/1

Y1 - 2005/12/1

N2 - The effects of vocal tract length (VTL) variation are often approximated by linear frequency warping of short-time spectra. Based on this relationship, we present a method for generating vocal tract length invariant features. These new features are computed as translation invariant, correlation-type features in a log-frequency domain. In phoneme classification experiments, their discrimination capabilities turned out to be considerably better than for Mel-frequency cepstral coefficients (MFCCs). The best results are obtained when VTL-invariant (VTLI) features and MFCCs are combined. The superiority of the combined feature set and its resilience to VTL variations is also shown for word recognition, using the TIDIGITS corpus and the HTK recognizer.

AB - The effects of vocal tract length (VTL) variation are often approximated by linear frequency warping of short-time spectra. Based on this relationship, we present a method for generating vocal tract length invariant features. These new features are computed as translation invariant, correlation-type features in a log-frequency domain. In phoneme classification experiments, their discrimination capabilities turned out to be considerably better than for Mel-frequency cepstral coefficients (MFCCs). The best results are obtained when VTL-invariant (VTLI) features and MFCCs are combined. The superiority of the combined feature set and its resilience to VTL variations is also shown for word recognition, using the TIDIGITS corpus and the HTK recognizer.

UR - https://www.scopus.com/record/display.uri?eid=2-s2.0-33846210086&origin=inward&txGid=77ccc47b766a274acebe9c4307d76f8e

U2 - 10.1109/ASRU.2005.1566473

DO - 10.1109/ASRU.2005.1566473

M3 - Conference Papers

SP - 33

EP - 37

T2 - 2005 IEEE Automatic Speech Recognition and Understanding Workshop

Y2 - 27 November 2005 through 1 December 2005

ER -

Vocal Tract Length Invariant Features for Automatic Speech Recognition

Abstract

Conference

UN SDGs

Access to Document

Other files and links

Fingerprint

Cite this