Improving Impulse Audio Source Separation using Generative Adversarial Networks for Phase Estimation

Author(s)
Piercy, Phoebe
Thumbnail
DownloadZip of audio files with readme (148.3Mb)
Terms of use
Music dataset comes from the The MUSDB18 corpus for music separation. Speech audio is taken from the Common Voice by Mozilla dataset (https://commonvoice.mozilla.org/en). Tactical sounds and commands are taken from the Military Sound Repository (https://militarysounds.org/home/browse/), the and the Modified Rhyme Test Audio Library (https://www.nist.gov/ctl/pscr/modified-rhyme-test-audio-library).
Metadata
Show full item record
Abstract
Audio files corresponding to the thesis on audio source separation using GAN phase estimation. They demonstrate various attempted methods to separate tactical impulse noise from speech and drum impulse noise from music, and show the mixed, clean and separated signals. These audio files are named in a numbering system according to the thesis section they correspond to, and should be listened to alongside the included plots. Also included are Modified Rhyme Test testing files.
Date issued
2021-04
URI
https://hdl.handle.net/1721.1/130559
Keywords
Audio Source Separation, Generative Adversarial Network, GAN, Noise removal, Time Frequency Masking, Tactical, Hearing Protection, Phase Estimation, Impulse, Intelligibility, Complex Masking
Collections