Statistics in Phonetics

Phonetics is the scientific study of speech production, perception, and acoustics. It is a data-rich field that integrates linguistics with physiology, psychology, physics, and machine learning. Researchers analyze diverse data types, including acoustic recordings, articulatory measurements (such as MRI or X-ray data), and neuroimaging. The field seeks to answer fundamental questions about how speech sounds are physically structured, how they vary across social groups and time, and how humans learn to produce and perceive them.

Historically, phonetic research relied on simple visual comparisons, t-tests, and ANOVA. However, the field has undergone a major methodological shift. As researchers began utilizing larger speech corpora—often facilitated by machine learning tools like forced alignment—the limitations of traditional models became apparent.

Modern phonetic analysis now favors:

• Linear Mixed-Effects Models: These are used to account for variability across different speakers and words, allowing for more robust inferences about the general population.
• Curve and Functional Data Analysis: Techniques like Generalized Additive Mixed Models (GAMMs) are increasingly used to model continuous speech data, such as the movement of the tongue or changes in pitch over time.
• Bayesian Methods: These are gaining traction for their flexibility in handling complex hierarchical data structures.

A significant portion of phonetic research involves transforming raw audio into interpretable features. Speech is typically recorded as a time series, but because speech is non-stationary, researchers often use frequency-domain analysis. Key representations include:

• Spectrograms: Time-frequency plots that visualize the energy of different sound frequencies.
• Formants: Resonant frequencies of the vocal tract (F1, F2, etc.) that are critical for identifying vowels and consonants.
• Source-Filter Model: A framework that models speech as a sound source (vocal fold vibration) passed through a filter (the vocal tract), which helps explain how different physical configurations produce distinct sounds.