Recently, the hierarchical extreme learning machine (HELM) model has been utilized for speech enhancement (SE) and demonstrated promising performance, especially when the amount of training data is limited and the system does not support heavy computations. Based on the success of audio-onlybased systems, termed AHELM, we propose a novel audio-visual HELM-based SE system, termed AVHELM that integrates the audio and visual information to confrontate the unseen nonstationery noise problem at low SNR levels to attain improved SE performance. The experimental results demonstrate that AVHELM can yield satisfactory enhancement performance with a limited amount of training data and outperforms AHELM in terms of three standardized objective measures under matched and mismatched testing conditions, confirming the effectiveness of incorporating visual information into the HELM-based SE system.

Audio-Visual Speech Enhancement using Hierarchical Extreme Learning Machine

Siniscalchi, Sabato Marco
Membro del Collaboration Group
;
2019

Abstract

Recently, the hierarchical extreme learning machine (HELM) model has been utilized for speech enhancement (SE) and demonstrated promising performance, especially when the amount of training data is limited and the system does not support heavy computations. Based on the success of audio-onlybased systems, termed AHELM, we propose a novel audio-visual HELM-based SE system, termed AVHELM that integrates the audio and visual information to confrontate the unseen nonstationery noise problem at low SNR levels to attain improved SE performance. The experimental results demonstrate that AVHELM can yield satisfactory enhancement performance with a limited amount of training data and outperforms AHELM in terms of three standardized objective measures under matched and mismatched testing conditions, confirming the effectiveness of incorporating visual information into the HELM-based SE system.
978-9-0827-9703-9
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11387/137368
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact