Show simple item record

dc.contributor.authorAnanthabhotla, Ishwarya
dc.contributor.authorEwert, Sebastian
dc.contributor.authorParadiso, Joseph A
dc.date.accessioned2021-12-15T14:32:25Z
dc.date.available2021-11-02T17:08:23Z
dc.date.available2021-12-15T14:32:25Z
dc.date.issued2019
dc.identifier.urihttps://hdl.handle.net/1721.1/137115.2
dc.description.abstract© 2019 Association for Computing Machinery. Generative audio models based on neural networks have led to considerable improvements across fields including speech enhancement, source separation, and text-to-speech synthesis. These systems are typically trained in a supervised fashion using simple element-wise ℓ1 or ℓ2 losses. However, because they do not capture properties of the human auditory system, such losses encourage modelling perceptually meaningless aspects of the output, wasting capacity and limiting performance. Additionally, while adversarial models have been employed to encourage outputs that are statistically indistinguishable from ground truth and have resulted in improvements in this regard, such losses do not need to explicitly model perception as their task; furthermore, training adversarial networks remains an unstable and slow process. In this work, we investigate an idea fundamentally rooted in psychoacoustics. We train a neural network to emulate an MP3 codec as a differentiable function. Feeding the output of a generative model through this MP3 function, we remove signal components that are perceptually irrelevant before computing a loss. To further stabilize gradient propagation, we employ intermediate layer outputs to define our loss, as found useful in image domain methods. Our experiments using an autoencoding task show an improvement over standard losses in listening tests, indicating the potential of psychoacoustically motivated models for audio generation.en_US
dc.language.isoen
dc.publisherAssociation for Computing Machinery (ACM)en_US
dc.relation.isversionof10.1145/3343031.3351148en_US
dc.rightsCreative Commons Attribution-Noncommercial-Share Alikeen_US
dc.rights.urihttp://creativecommons.org/licenses/by-nc-sa/4.0/en_US
dc.sourceMIT web domainen_US
dc.titleTowards a Perceptual Loss: Using a Neural Network Codec Approximation as a Loss for Generative Audio Modelsen_US
dc.typeArticleen_US
dc.identifier.citationAnanthabhotla, Ishwarya, Ewert, Sebastian and Paradiso, Joseph A. 2019. "Towards a Perceptual Loss: Using a Neural Network Codec Approximation as a Loss for Generative Audio Models." MM 2019 - Proceedings of the 27th ACM International Conference on Multimedia.en_US
dc.contributor.departmentMassachusetts Institute of Technology. Media Laboratoryen_US
dc.contributor.departmentProgram in Media Arts and Sciences (Massachusetts Institute of Technology)en_US
dc.relation.journalMM 2019 - Proceedings of the 27th ACM International Conference on Multimediaen_US
dc.eprint.versionAuthor's final manuscripten_US
dc.type.urihttp://purl.org/eprint/type/ConferencePaperen_US
eprint.statushttp://purl.org/eprint/status/NonPeerRevieweden_US
dc.date.updated2021-06-25T17:49:20Z
dspace.orderedauthorsAnanthabhotla, I; Ewert, S; Paradiso, JAen_US
dspace.date.submission2021-06-25T17:49:21Z
mit.licenseOPEN_ACCESS_POLICY
mit.metadata.statusPublication Information Neededen_US


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record

VersionItemDateSummary

*Selected version