Burin Naowarat
burin.naowarat@ed.ac.uk
I wonder how neural networks process complex sensory input to make accurate predictions. I am currently exploring this question through interpretability research on speech models, studying geometric properties of hidden structures in their representations.
I’m a PhD student at the Centre for Speech Technology Research (CSTR) in the School of Informatics at the University of Edinburgh, supervised by Sharon Goldwater and Hao Tang. My current work studies the low-dimensional geometry of representations in deep speech models to understand how they simultaneously represent different aspects of speech, such as what is said, who is speaking, and how speech unfolds over time.
Before starting my PhD, I completed a research master’s degree focused on ASR at Chulalongkorn University, supervised by Ekapol Chuangsuwanich. During my master’s, I interned with Amazon Alexa in Cambridge, UK. I later worked as a language modelling engineer for on-device ASR systems at Cerence.
Publications
View on Google Scholar ↗-
Deep Speech Representations Track Word Timing along a Low-Dimensional Circular Trajectory
, Hao Tang, and Sharon Goldwater.
-
A framework for analyzing concept representations in neural models
, Hao Tang, and Sharon Goldwater.
-
Effective Training of Attention-based Contextual Biasing Adapters with Synthetic Audio for Personalised ASR
, Philip Harding, Pasquale D’Alterio, Sibo Tong, and Bashar Awwad Shiekh Hasan.
Interspeech, 2023. Paper ↗
-
Word-level Confidence Estimation for CTC Models
, Thananchai Kongthaworn, and Ekapol Chuangsuwanich.
-
Thai Dialect Corpus and Transfer-based Curriculum Learning Investigation for Dialect Automatic Speech Recognition
Artit Suwanbandit, , Orathai Sangpetch, and Ekapol Chuangsuwanich.
-
Context Conditioning via Surrounding Predictions for Non-Recurrent CTC Models
, Chawan Piansaddhayanon, and Ekapol Chuangsuwanich.
IEEE Access, 2023. Paper ↗
-
Spectral and Latent Speech Representation Distortion for TTS Evaluation
Thananchai Kongthaworn, , and Ekapol Chuangsuwanich.
Interspeech, 2021. Paper ↗
-
Set Prediction in the Latent Space
Konpat Preechakul, Chawan Piansaddhayanon, , Tirasan Khandhawit, Sira Sriswasdi, and Ekapol Chuangsuwanich.
-
Reducing Spelling Inconsistencies in Code-Switching ASR using Contextualized CTC Loss
, Thananchai Kongthaworn, Korrawe Karunratanakul, Sheng Hui Wu, and Ekapol Chuangsuwanich.
ICASSP, 2021. Paper ↗