DR. GEORGE CLOSE

Machine Learning Researcher and Speech Data Scientist

Me IRL

Email: george.close@connex.ai

I am a machine learning researcher with a focus on speech processing tasks. I am currently working as a Speech Data Scientist at ConnexAI.
I hold a PhD in Computer Science from the University of Sheffield, and a BSc in Computer Science from Cardiff University.
My research interests include neural speech enhancement, automatic speech recognition, speech quality assessment and human perception of audio. I an an expert in a number of deep learning frameworks including PyTorch and TensorFlow, and have extensive experience with GPU clusters and Linux systems.

Interests

  • Neural Speech Enhancement / Noise Reduction
  • Psychoacoustically motivated approaches
  • Automatic Speech Recognition (ASR)
  • Speech Quality / Intelligibility assessment metrics and prediction
  • Human perception of audio
  • Neural systems for hearing aids and other low power devices
  • Self Supervised Speech Representations / Foundational models
  • Deep Fake Detection / Adversial attacks on neural networks.

Technical Skills

  • Python - PyTorch, Tensorflow, SciPy, SpeechBrain
  • HuggingFace / OpenAI API
  • Git / GitHub
  • High performance computing / GPU clusters
  • CuDA
  • MATLAB
  • Linux / BASH scripting
  • C++, Java, Haskell

Qualifications & Work Experience

  • Speech Data Scientist @ ConnexAI
    Manchester, UK
    November 2024 - Present

  • Yamaha Research and Development (Internship)
    Hammamatsu, Japan
    May 2024 - August 2024

  • PhD Computer Science + GTA Teaching Assisstant
    Thesis Title: Perceptually Motivated Speech Enhancement
    University of Sheffield, UK
    October 2020 - January 2025

  • BSc Computer Science (First Class Honours)
    Thesis Title: Majel - Voice control for Command Line Interfaces
    Cardiff University, UK
    August 2017- August 2020

Papers and Publications

I am an author on 12 papers, of which 9 I am first author.

  • Hallucination in Perceptual Metric-Driven Speech Enhancement Networks
    arXiv | EUSIPCO 2024 | Listening Test Examples
    George Close, Thomas Hain and Stefan Goetze

  • SSSR In Loss Functions For Hearing Aid Speech Enhancement
    EUSIPCO 2024
    Robert Sutherland, George Close, Thomas Hain, Stefan Goetze and Jon Barker

  • Transcription-Free Fine-Tuning of Speech Separation Models for Noisy and Reverberant Multi-Speaker Automatic Speech Recognition
    arXiv | Interspeech 2024
    William Ravenscroft, George Close, Stefan Goetze, Thomas Hain, Mohammad Soleymanpour, Anurag Chowdhury and Mark C. Fuhs

  • Non-Intrusive Speech Intelligibility Prediction for Hearing-Impaired Users using Intermediate ASR Features and Human Memory Models
    Techincal Report | arXiv | ICASSP 2024
    🏆 2nd Place in Clarity Prediction Challenge 2
    Rhiannon Mogridge, George Close, Robert Sutherland, Thomas Hain, Jon Barker, Stefan Goetze and Anton Ragni

  • Multi-CMGAN+/+: Leveraging Multi-Objective Speech Quality Metric Prediction for Speech Enhancement
    arXiv | ICASSP 2024 | GitHub
    George Close, Thomas Hain and Stefan Goetze

  • CMGAN+/+: The University of Sheffield CHiME-7 UDASE Challenge Speech Enhancement System
    Technical Report
    🏆 Entry to CHiME 2024 UDASE Task

    George Close, William Ravenscroft, Thomas Hain and Stefan Goetze

  • Non Intrusive Intelligibility Predictor for Hearing Impaired Individuals using Self Supervised Speech Representations
    arXiv | SPARKS Workshop 2023
    George Close, Thomas Hain and Stefan Goetze

  • The Effect of Spoken Language on Speech Enhancement using Self-Supervised Speech Representation Loss Functions
    arXiv | WASPAA 2023 | GitHub
    George Close, Thomas Hain, Stefan Goetze

  • Perceive and predict: Self-Supervised Speech Representation Based Loss Functions for Speech Enhancement
    arXiv | ICASSP 2023 | Audio Examples
    George Close, William Ravenscroft, Thomas Hain and Stefan Goetze

  • PAMGAN+/-: Improving Phase Aware Speech Enhancement Performance via Expanded Discriminator Training
    154th AES Convention Europe 2023
    🏆 WINNER: STUDENT TECHNICAL PAPER AWARD
    George Close, Thomas Hain and Stefan Goetze

  • Non-intrusive Speech Intelligibility Metric Prediction for Hearing Impaired Individuals - Clarity Prediction Challege 1
    Interspeech 2022
    George Close, Samuel Hollands, Thomas Hain and Stefan Goetze

  • MetricGAN+/- Increasing Robustness of Noise Reduction on Unseen Data
    arXiv | EUSIPCO 2022 | Audio Examples
    George Close, Thomas Hain and Stefan Goetze

Talks and Presentations

My github My twitter My LinkIn My Google Scholar