Ridwan Bello

PhD Student in Machine Learning & AI

Research Overview

My PhD research focuses on developing speech recognition models that address the challenges of low-resource languages, accented speech, and fairness in AI. I have worked on a range of projects including accented speech classification, bias detection in ASR systems, and multimodal learning approaches for African languages. My work aims to contribute to both academic advancements and real-world applications of machine learning in speech technologies.

Detailed Project Pages

Accent Classification with Wav2Vec2

A project aimed at building a robust accent classification system using MFCC features and Wav2Vec2 embeddings.

  • Role: Led data curation, model development, and evaluation.
  • Technologies: PyTorch, MFCC, Wav2Vec2, Scikit-learn.
  • Outcomes: Achieved 82% classification accuracy; contributed to a published paper.
View Detailed Page →

Bias & Fairness in Speech Recognition

An in-depth analysis of bias in ASR systems and proposed fairness-aware training pipelines.

  • Role: Designed experiments and fairness metrics evaluation.
  • Technologies: TensorFlow, Fairlearn, Hugging Face Transformers.
  • Outcomes: Presented findings at African AI Conference 2024.
View Detailed Page →

Multi-Task Learning for African-Accented English Speech

This project focuses on building a multi-task learning model that simultaneously performs accent recognition, gender recognition, and age recognition from African-accented English speech, enhancing ASR performance and demographic profiling.

  • Role: Led dataset preparation, model design, and multi-task training strategy.
  • Technologies: PyTorch, Hugging Face Transformers, Wav2Vec2, FastAPI.
  • Outcomes: Early results show strong performance across all tasks, with improved model generalization and deeper insights into speaker demographics.
View Detailed Page →

Racial Bias in Facial Expression Recognition Datasets

This project investigates the presence of racial bias in popular facial expression recognition (FER) datasets and its impact on model fairness, aiming to improve equity in computer vision systems.

  • Role: Conducted dataset audits, bias quantification, and model fairness evaluation.
  • Technologies: Python, OpenCV, TensorFlow/Keras, Scikit-learn.
  • Outcomes: Identified significant disparities in FER model performance across racial groups; proposed dataset balancing and models for marginalized groups.
View Detailed Page →