Skip to Main content Skip to Navigation
New interface
Conference papers

Nonnegative Feature Learning Methods for Acoustic Scene Classification

Victor Bisot 1, 2 Romain Serizel 3 Slim Essid 2, 1 Gael Richard 2, 1 
3 MULTISPEECH - Speech Modeling for Facilitating Oral-Based Communication
Inria Nancy - Grand Est, LORIA - NLPKD - Department of Natural Language Processing & Knowledge Discovery
Abstract : This paper introduces improvements to nonnegative feature learning-based methods for acoustic scene classification. We start by introducing modifications to the task-driven nonnegative matrix factorization algorithm. The proposed adapted scaling algorithm improves the generalization capability of task-driven nonneg-ative matrix factorization for the task. We then propose to exploit simple deep neural network architecture to classify both low level time-frequency representations and unsupervised nonnegative matrix factorization activation features independently. Moreover, we also propose a deep neural network architecture that exploits jointly unsupervised nonnegative matrix factorization activation features and low-level time frequency representations as inputs. Finally, we present a fusion of proposed systems in order to further improve performance. The resulting systems are our submission for the task 1 of the DCASE 2017 challenge.
Document type :
Conference papers
Complete list of metadata

Cited literature [24 references]  Display  Hide  Download
Contributor : Romain Serizel Connect in order to contact the contributor
Submitted on : Thursday, November 16, 2017 - 5:37:24 PM
Last modification on : Friday, January 21, 2022 - 3:09:14 AM
Long-term archiving on: : Saturday, February 17, 2018 - 5:05:23 PM


Files produced by the author(s)


  • HAL Id : hal-01636627, version 1


Victor Bisot, Romain Serizel, Slim Essid, Gael Richard. Nonnegative Feature Learning Methods for Acoustic Scene Classification. DCASE 2017 - Workshop on Detection and Classification of Acoustic Scenes and Events, Nov 2017, Munich, Germany. ⟨hal-01636627⟩



Record views


Files downloads