William Ravenscroft
Audio and speech machine learning researcher
Summary
Work
Education
Projects
Writing
Whilter: A Whisper-based Data Filter for "In-the-Wild" Speech Corpora Using Utterance-level Multi-Task Classification
January 1, 2025A data-filtering approach leveraging Whisper for quality filtering of in-the-wild speech corpora (Interspeech 2025 / arXiv preprint).
Speech Separation in Noisy Reverberant Acoustic Environments (PhD Thesis)
January 1, 2024Doctoral thesis investigating assumptions, architectures and training strategies for speech separation and enhancement, with aims to improve generalisation, robustness and computational efficiency.
Transcription-Free Fine-Tuning of Speech Separation Models for Noisy and Reverberant Multi-Speaker Automatic Speech Recognition
January 1, 2024Work on fine-tuning speech separation models without full transcriptions to improve ASR in noisy, reverberant, multi-speaker scenarios; accepted at Interspeech 2024.
On Data Sampling Strategies for Training Neural Network Speech Separation Models
January 1, 2023Study analysing the impact of training signal length and data sampling strategies on speech separation model performance and training efficiency.
Deformable temporal convolutional networks for monaural noisy reverberant speech separation
January 1, 2022Paper presenting deformable temporal convolutional network approaches for speech separation in noisy and reverberant conditions.