analyzing pediatric voices

I took a linguistics class and got hooked on learning about human speech and the science of communication. I joined my linguistics professors lab where I helped out in a study by building an automated speech-transcription and phonetic-alignment pipeline for child speech data using Praat, whisper.cpp, and the Montreal Forced Aligner (MFA). These technologies use probability modeling to find phonemes from MRI voice recordings. I automated this workflow so that it would flag words that appeared out of dictionary- find the phonemes for it and ask for human approval before continuing.

I am currently working on an extension to adapt pretrained acoustic models to improve phoneme-level alignment accuracy for pediatric voice samples.