Massa Baali

I'm a PhD researcher at Carnegie Mellon University's Language Technologies Institute, advised by Professor Bhiksha Raj. My research centers on speech foundation models, explainable speaker verification, and speech language models — systems that don't just identify a voice, but describe it in language and remain reliable under real-world conditions.

In my first year of the PhD I led four first-author projects — CoLMbo, CoLMbo-SV, SVeritas, and CAARMA — and produced six research papers, with work appearing at ASRU, SLT, and EMNLP. The thread connecting them is a belief that voice technology should be both trustworthy and interpretable: a verification model should be able to explain its decision, and a speaker model should be able to say what it hears.

Before CMU, I spent two years as an AI researcher at MBZUAI, where I was a core contributor to JAIS, a 13-billion-parameter Arabic LLM trained on 395 billion tokens, and built end-to-end check-worthiness detection for political debates. Earlier, at Kanari AI, I developed dialectal Arabic speech recognition that outperformed Google, Microsoft, and Amazon, and helped introduce text-to-speech in collaboration with QCRI and CMU. I began my career as a startup founder, building an AI-driven recruitment platform that was accepted into the Founder Institute and MIT's entrepreneurship programs.

Beyond research, I care deeply about teaching and mentorship — leading recitations for CMU's flagship deep learning course, mentoring students at the LTI, and earlier teaching math, English, and computer science to refugees through the Jusoor NGO.