Hi, Salut, Hallo, ဟိုင်း!

Having lived in 8 countries by 21, Where are you from? is a much harder question for me to answer than Where are you living? For the easier question, the current answer is: Oakland, California. For the harder question, a simplified answer is: Australia (I do hold an Australian passport after all).

Similarly, Where do you work? is an easy one. Currently, I’m a Staff Engineer at rime, where I help build conversational voice AI systems. Before that I was a grad student in Linguistics at Stanford (advised by Dan Jurafsky), working on improving access to untranscribed speech corpora using AI.

What do you work on? is a harder question. A simplified answer is: speech and language processing (I did get some Linguistics degrees after all). A longer but somewhat abstract answer is that I enjoy working on projects that are simply impossible to tackle alone and without technological assistance, thinking through all the moving parts and many participants, and architecting the systems and refining the processes that glue everything and everyone together. For more concrete answers, have a gander at some of the projects below.

Selected Writing

Selected Work

Speech processing

Acoustic Token Distribution Similarity
A similarity metric that predicts which higher-resource language can best “donate” data to improve speech recognition for a low-resource one

Speech tools for language revival
A privacy-preserving pipeline that auto-transcribes the English commentary in archival recordings, so communities can triage which audio to review and annotate

Searching untranscribed speech
Helping communities and linguists locate words in collections of endangered language audio recordings without the need for time-consuming transcriptions

Phonetics and phonology

Clustering of Kaytetye vowels
A phonetic and phonological analysis of the Kaytetye vowel system, arguing it contrasts four vowels rather than the two or three previously proposed

Text-setting in Kaytetye
A computational analysis of how a Kaytetye ceremonial song tradition fits spoken words to its rhythm by deleting, adding, and moving syllables

Reproducible collaborative annotation
Brings reproducible version control to collaborative speech annotation through a graphical client, so annotators need no command-line skills

Digital and print dictionaries

Warlpiri Encyclopaedic Dictionary — book cover
Collaborative data validation and conversion pipeline to help produce the largest print dictionary of an Australian Aboriginal language (1400+ pages)

Yerrampe: Kaytetye Multimedia Dictionary — landing page
Offline multimedia dictionary of ~4000 Kaytetye headwords, built as static HTML pages with text and studio-recorded audio generated from a plain-text database