Effect of Environment in Speech Quality Assessment in Crowdsourcing - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2020

Effect of Environment in Speech Quality Assessment in Crowdsourcing

Résumé

The quality of the speech signal is of main importance as it influences the user experience of interactive systems such as: personal assistants, telephony calls, virtual conversational agents, and others. Speech quality studies have been traditionally conducted in constrained laboratory rooms with professional audio equipment. Nowadays, crowdsourcing (CS) represent a valid alternative for the rapid assessment of large speech databases. However, the question remains regarding the influence of the listeners' environmental background noise, and context. This paper compares two speech quality studies that were conducted in the laboratory following international standards, i.e ITU-T Rec. P.800 (Lab-traditional) and ITU-T Rec. P.808 (CS-simulated), respectively. During the test, listeners were exposed to background noise at different levels, and web audio recordings were also collected. We found that there was a statistically significant interaction between environment (Lab and CS) and level of noise, on the speech quality ratings provided by the listeners when assessing only one conditions out the 15 that were under test. Additionally, we trained multiple state-of-the-art classifiers with the web audio recordings to evaluate the validity of this data to detect the participants' environment conditions.
Fichier principal
Vignette du fichier
000995.pdf (221.41 Ko) Télécharger le fichier
Origine Fichiers éditeurs autorisés sur une archive ouverte

Dates et versions

hal-03235391 , version 1 (27-05-2021)

Identifiants

Citer

Rafael Zequeira Jiménez, Babak Naderi, Sebastian Möller. Effect of Environment in Speech Quality Assessment in Crowdsourcing. Forum Acusticum, Dec 2020, Lyon, France. pp.3033-3037, ⟨10.48465/fa.2020.0995⟩. ⟨hal-03235391⟩

Collections

FA2020
46 Consultations
33 Téléchargements

Altmetric

Partager

Gmail Mastodon Facebook X LinkedIn More