Static, dynamic, and relational properties in vowel perception

Area of Science:

• Acoustic Phonetics
• Speech Perception
• Machine Learning in Auditory Analysis

Background:

• Vowel perception is influenced by speaker variability and consonantal context, traditionally analyzed via formant overlap in F1 x F2 space.
• Machine recognition approaches are compared with empirical speech perception data for English vowels.
• Understanding factors affecting vowel perception is crucial for both linguistic theory and auditory technology development.

Purpose of the Study:

• To review theories and empirical findings on English vowel perception.
• To compare data analytic "machine recognition" approaches with speech perception experiments.
• To investigate the influence of speaker normalization and consonantal context on vowel perception.

Main Methods:

• Review of existing theories and empirical findings on English vowel perception.
• Presentation of two new experiments investigating speaker normalization and consonantal context effects.
• Analysis of vowel perception data using acoustic measurements (formant frequencies, duration) and perceptual judgments.
• Evaluation of intrinsic and extrinsic speaker normalization techniques.
• Extension of previous findings on consonantal context effects to new phonetic contexts.

Main Results:

• New experimental evidence supports the role of both intrinsic and extrinsic speaker normalization in vowel perception.
• Consonantal context effects on vowel perception were confirmed and extended to /dVd/ and /bVb/ contexts.
• Vowel-inherent dynamic properties like duration and diphthongization also reliably influence perception.
• Both machine recognition and human perception are affected by speaker differences and consonantal context.

Conclusions:

• Speaker normalization and consonantal context are significant factors influencing English vowel perception.
• Vowel-inherent dynamic properties also play a reliable role in how vowels are perceived.
• Further research with complex perceptual experiments and quantitative pattern recognition models is needed to fully resolve the complexities of vowel perception.