Acoustic cues are defined as physical features that profile various speech production events. A cue of particular interest is the Spectral Burst (SB), which can be leveraged to predict the place of articulation of a consonant sound in the vocal tract. In turn, accurately classifying Spectral Bursts can create more robust Automatic Speech Recognition (ASR) that is reliable cross-linguistically. In this project, I will develop a statistical machine learning model that detects Spectral Bursts in phonologically complex settings. Then, I will test this model to see if it processes speech in accordance with known human perceptual benchmarks. This requires understanding both the low-level phonetic decomposition and high-level probabilistic approaches necessary to analyze speech psycholinguistically. This model can then be applied to fields such as speech disorder diagnostics, emotion recognition, and interactive voice systems.

I am participating in SuperUROP to experience end-to-end research in a field of interest, Speech Processing. My project will utilize theory from some of my favorite courses at MIT: 9.35 (Perception), 6.390 (ML), and 24.900 (Linguistics), alongside skills and interests such as Signal Processing and Acoustic Phonetics. I ultimately hope to produce research that is both computationally robust and aligned with underlying psycholinguistic theory.