Quantifying robustness of the /t/-/k/ contrast using a single, static spectral feature

Area of Science:

• Phonetics
• Speech Acoustics
• Linguistics

Background:

• Accurate classification of speech sounds like /t/ and /k/ is crucial for understanding speech production and perception.
• Previous research often relies on complex dynamic spectral features, necessitating simpler methods for broader application.

Purpose of the Study:

• To investigate if a single, static spectral feature can reliably differentiate between word-initial /t/ and /k/ productions.
• To compare the efficacy of centroid and peak frequency in classifying these consonants across speakers and vowel contexts.

Main Methods:

• Analysis of acoustic and psychoacoustic spectra from /t/ and /k/ tokens produced by 21 adult speakers.
• Calculation and comparison of centroid and peak frequency as potential differentiating features.
• Evaluation of feature performance relative to speaker and vowel context.

Main Results:

• A single static spectral feature, centroid frequency, demonstrated accurate classification of /t/ and /k/ productions.
• Both acoustic and psychoacoustic centroid frequency measures were effective in differentiation.
• The classification accuracy was maintained across different speakers and vowel contexts.

Conclusions:

• Static spectral features, specifically centroid frequency, offer a robust and simplified method for distinguishing /t/ and /k/.
• This finding has implications for speech analysis, speech synthesis, and the development of diagnostic tools for speech disorders.