M.A. Thesis · Indiana University
Acoustic Phonetics
An acoustic study of F1/F2 formant transfer in L1 English learners of L2 Turkish, built with Praat and Audacity, advised by
Dr. Öner Özçelik. Continued today in Timbre.
Mole Analysis with Deep Adam Optimized Inception Network
Machine Learning · CNN·Team of 4
Convolutional neural network (InceptionV3 with Adam, built in TensorFlow and Keras) that sorts dermatoscopic skin lesion images into 7 diagnostic classes, 3 malignant and 4 benign, across the 10,015 image HAM10000 dataset. We benchmarked 9 architecture and optimizer combinations and tuned class weighting to prioritize melanoma recall, since false negatives are the most costly error. Deployed as a Streamlit web app.
Acoustic Phonetics Lab
Web Audio · DSP
Browser based vowel space analyzer and research writeup, a direct extension of my thesis work above: a cardinal vowel chart and reference formant data (F1/F2/F3) across 41 languages, with live microphone extraction as the next step.
Real time AI Sign Language Interpreter
Computer Vision · ML
In browser ASL fingerspelling recognition using MediaPipe hand tracking and an ONNX model (39 classes, ~99% validation accuracy), with a Claude assisted text-to-sign mode. ISL is defined in the UI but has no trained model behind it yet.