Michael Saxon

Ph.D. Student, ECE+NLP
UC Santa Barbara

Google Scholar
Updated January 26, 2019

Bioinspired Deep Speech Processing and Synthesis

2019 — present

My master's thesis project involving adapting neuromuscular modelling to drive better deep speech synthesis and processing.

Affective Computing

2018 — present

Created novel Word Pair Convolutional Model for classification of speaker agency and sociality of a self-description of happy moments for the CL-Aff shared task that achieved runner-up highest classification accuracy on the test set, and over 91% accuracy on cross validation evaluation.

Assisting in user study on the impact of audio/facial expression entrainment and conflict in emotion identification for visually impaired participants recieving haptic facial expression representations.

Principled LiDAR Processing

2017 — present

Assisting in the collection of a pedestrian crowd "millions of LiDAR frames" dataset to facilitate a "predict next n LiDAR frames given previous m" task, and developing neural network architectures that can process the data.

Evaluated the performance of low-cost 16-segment solid state LiDAR units from Velodyne within an integrated video/distance frame paradigm, work in review for SPIE Journal.

Disordered Speech Characterization

2017 — 2018

Created ASR-based objective measures of hypernasality in speech leveraging common nasal/plosive co-located "cognates" for use in larger nasality classification systems that are disorder-agnostic. Work under review for ICASSP 2019

Aided in the integration of multiple objective measures including an acoustic feature and Goodness of Pronunciation for disorder-agnositc nasalit y classification. Work will be submitted to IEEE SPS.

Strain Measurement Apparatus Control System

2014 — 2015

Created integrated control and measurement system for sub micron-scale semiconductor optical strain measurement in LabVIEW.

© 2017-2020 Michael Saxon. Source available on GitHub.    [Icon Attribution]