Exploring the Relationship Between Bias and Speech Acoustics in Automatic Speech Recognition Systems
This repository contains the code used for my Bachelor thesis conducted for Research Project 2024 at the TU Delft. The topic of the project was "Exploring the Relationship Between Bias and Speech Acoustics in Automatic Speech Recognition Systems".
fetch.py
: fetches the speech files from the datasetgroup.sh
: gathers the metadata about the speaker groupscalculate_bias.sh
: calculates the bias based on the word error ratesfeaturize.py
: computes the acoustic embeddings for speech filesdistance.py
: computes the distance between the acoustic embeddingsplot.py
: plots scatter plots for bias against the acoustic distancerequirements.txt
: lists the libraries and their versions used for the project
The scripts assume the existence of data that is not present in this repository for licensing reasons. The code should be adapted to the dataset structure.