A New Vowel Normalization for Sociophonetics

W.J. Heeringa; H. Van de Velde

doi:10.21437/Interspeech.2021-1846

Abstract

Several studies have shown that in sociophonetic research Lobanov’s speaker normalization method outperforms other methods for normalizing vowel formants of speakers. An advantage of Lobanov’s method compared to the method that was introduced by Watt & Fabricius in 2002 is that it is independent of the shape of the vowel space area, and also normalizes to the dispersion of the vowels. However, it does depend on the distribution of the vowels within the vowel space. When using Lobanov normalization the formant values are converted to z-scores. We present a method where the µ in the z-score formula is replaced by the center of the convex hull that encloses the vowels, and the σ is obtained on the basis of the points that constitute the convex hull. When normalizing measurements of two real data sets, and of a series of randomly generated data sets, we found that our method improved in matching vowel spaces in size and overlap.

Original language	English
Pages	4024-4028
DOIs	https://doi.org/10.21437/Interspeech.2021-1846
Publication status	Published - Dec 2021
Event	Interspeech 2021 - Brno, Czech Republic Duration: 30 Aug 2021 → 03 Sept 2021 https://www.interspeech2021.org/

Conference

Conference	Interspeech 2021
Country/Territory	Czech Republic
City	Brno
Period	30/08/2021 → 03/09/2021
Internet address	https://www.interspeech2021.org/

Keywords

vowel formants
vowel space
normalization
sociophonetics

Access to Document

10.21437/Interspeech.2021-1846Licence: CC BY-NC-SA

https://www.isca-speech.org/archive/interspeech_2021/heeringa21_interspeech.html

A New Vowel Normalization for Sociophonetics
W.J. Heeringa (Speaker) & H. Van de Velde (Speaker)
03 Sept 2021
Activity: Talk or presentation › Academic

File

Cite this

@conference{296c71ebfb314991859748980f51cf7b,

title = "A New Vowel Normalization for Sociophonetics",

abstract = "Several studies have shown that in sociophonetic research Lobanov{\textquoteright}s speaker normalization method outperforms other methods for normalizing vowel formants of speakers. An advantage of Lobanov{\textquoteright}s method compared to the method that was introduced by Watt & Fabricius in 2002 is that it is independent of the shape of the vowel space area, and also normalizes to the dispersion of the vowels. However, it does depend on the distribution of the vowels within the vowel space. When using Lobanov normalization the formant values are converted to z-scores. We present a method where the µ in the z-score formula is replaced by the center of the convex hull that encloses the vowels, and the σ is obtained on the basis of the points that constitute the convex hull. When normalizing measurements of two real data sets, and of a series of randomly generated data sets, we found that our method improved in matching vowel spaces in size and overlap.",

keywords = "vowel formants, vowel space, normalization, sociophonetics",

author = "W.J. Heeringa and {Van de Velde}, H.",

year = "2021",

month = dec,

doi = "10.21437/Interspeech.2021-1846",

language = "English",

pages = "4024--4028",

note = "Interspeech 2021 ; Conference date: 30-08-2021 Through 03-09-2021",

url = "https://www.interspeech2021.org/",

}

TY - CONF

T1 - A New Vowel Normalization for Sociophonetics

AU - Heeringa, W.J.

AU - Van de Velde, H.

PY - 2021/12

Y1 - 2021/12

N2 - Several studies have shown that in sociophonetic research Lobanov’s speaker normalization method outperforms other methods for normalizing vowel formants of speakers. An advantage of Lobanov’s method compared to the method that was introduced by Watt & Fabricius in 2002 is that it is independent of the shape of the vowel space area, and also normalizes to the dispersion of the vowels. However, it does depend on the distribution of the vowels within the vowel space. When using Lobanov normalization the formant values are converted to z-scores. We present a method where the µ in the z-score formula is replaced by the center of the convex hull that encloses the vowels, and the σ is obtained on the basis of the points that constitute the convex hull. When normalizing measurements of two real data sets, and of a series of randomly generated data sets, we found that our method improved in matching vowel spaces in size and overlap.

AB - Several studies have shown that in sociophonetic research Lobanov’s speaker normalization method outperforms other methods for normalizing vowel formants of speakers. An advantage of Lobanov’s method compared to the method that was introduced by Watt & Fabricius in 2002 is that it is independent of the shape of the vowel space area, and also normalizes to the dispersion of the vowels. However, it does depend on the distribution of the vowels within the vowel space. When using Lobanov normalization the formant values are converted to z-scores. We present a method where the µ in the z-score formula is replaced by the center of the convex hull that encloses the vowels, and the σ is obtained on the basis of the points that constitute the convex hull. When normalizing measurements of two real data sets, and of a series of randomly generated data sets, we found that our method improved in matching vowel spaces in size and overlap.

KW - vowel formants

KW - vowel space

KW - normalization

KW - sociophonetics

U2 - 10.21437/Interspeech.2021-1846

DO - 10.21437/Interspeech.2021-1846

M3 - Abstract

SP - 4024

EP - 4028

T2 - Interspeech 2021

Y2 - 30 August 2021 through 3 September 2021

ER -

A New Vowel Normalization for Sociophonetics

Abstract

Conference

Keywords

Access to Document

Fingerprint

Activities

A New Vowel Normalization for Sociophonetics

Cite this