The Entanglement of Dialectal Variation and Speaker Normalization

Area of Science:

• Sociolinguistics
• Phonetics
• Computational Linguistics

Background:

• Speaker normalization is crucial for analyzing acoustic variation in speech.
• Different normalization algorithms exist, based on distinct theoretical underpinnings.
• Sociolinguistic variation, such as dialectal identity, can be masked by normalization procedures.

Purpose of the Study:

• To investigate the impact of two distinct speaker normalization procedures on sociolinguistic variation in American English.
• To determine if normalization methods preserve or eliminate dialectal identity markers in vowel formants.
• To compare the effectiveness of Lobanov and Labov ANAE normalization in acoustic analyses of heritage speakers.

Main Methods:

• Analysis of vowel formants from 88 American English speakers with Finnish and Italian heritage.
• Application of two normalization procedures: Lobanov and Labov ANAE.
• Comparison of normalized formant data to identify preservation or elimination of sociolinguistic variation.

Main Results:

• The Labov ANAE normalization method revealed significant vowel space shifts correlated with speaker heritage.
• The Lobanov normalization method appeared to eliminate these heritage-based sociolinguistic variations.
• Differences in results highlight how normalization algorithms handle formant relationships and acoustic variation.

Conclusions:

• The Lobanov normalization procedure may over-normalize acoustic data, removing sociolinguistically relevant information.
• The Labov ANAE method demonstrates greater utility in preserving dialectal identity markers in acoustic analyses.
• The choice of normalization technique significantly influences the interpretation of sociolinguistic variation in speech data.