Cristin-resultat-ID: 1488349
Sist endret: 21. februar 2018, 15:02
NVI-rapporteringsår: 2017
Resultat
Vitenskapelig artikkel
2017

Subjective Intelligibility of Deep Neural Network-Based Speech Enhancement

Bidragsytere:
  • Femke B. Gelderblom
  • Tron Vedul Tronstad og
  • Erlend Magnus Viggen

Tidsskrift

Interspeech (USB)
ISSN 2308-457X
NVI-nivå 1

Om resultatet

Vitenskapelig artikkel
Publiseringsår: 2017
Volum: 2017-August
Sider: 1968 - 1972
Open Access

Importkilder

Scopus-ID: 2-s2.0-85039168092

Beskrivelse Beskrivelse

Tittel

Subjective Intelligibility of Deep Neural Network-Based Speech Enhancement

Sammendrag

Recent literature indicates increasing interest in deep neural networks for use in speech enhancement systems. Currently, these systems are mostly evaluated through objective measures of speech quality and/or intelligibility. Subjective intelligibility evaluations of these systems have so far not been reported. In this paper we report the results of a speech recognition test with 15 participants, where the participants were asked to pick out words in background noise before and after enhancement using a common deep neural network approach. We found that, although the objective measure STOI predicts that intelligibility should improve or at the very least stay the same, the speech recognition threshold, which is a measure of intelligibility, deteriorated by 4 dB. These results indicate that STOI is not a good predictor for the subjective intelligibility of deep neural network-based speech enhancement systems. We also found that the postprocessing technique of global variance normalisation does not significantly affect subjective intelligibility.

Bidragsytere

Femke Gelderblom

Bidragsyterens navn vises på dette resultatet som Femke B. Gelderblom
  • Tilknyttet:
    Forfatter
    ved Connectivity Technologies and Platforms ved SINTEF AS

Tron Vedul Tronstad

  • Tilknyttet:
    Forfatter
    ved Connectivity Technologies and Platforms ved SINTEF AS

Erlend Magnus Viggen

  • Tilknyttet:
    Forfatter
    ved Connectivity Technologies and Platforms ved SINTEF AS
1 - 3 av 3