Lower level acoustic characteristics are intrinsically linked with the phonemes comprising a language, and have distinct acoustic patterns (which can be analyzed using spectrogram data) that are especially useful for phonological classification models. I will be working on furthering an existing SB (spectral burst/noise) detection module to detect the types of turbulence noise that are characteristic of obstruent consonants, specifically to distinguish between the spectral burst noise of the /s/ and /sh/ phonemes. Our approach is to implement and test a GMM (Gaussian mixture model)-based SB detection module that is able to replicate results from human listening experiments for these sounds.

I hope to learn more about high quality research practices as well as effective communication techniques. Having analyzed speech acoustics in English phonemes using spectrograms, I am excited to learn more about how humans can distinguish between various sounds.