Towards Exploring Linguistic Variation in ASR Errors: Paradigm & Tool for Perceptual experiments - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2011

Towards Exploring Linguistic Variation in ASR Errors: Paradigm & Tool for Perceptual experiments

Résumé

It is well-known that human listeners significantly outperform machines when it comes to transcribing speech. This paper presents a paradigm for perceptual experiments that aims to increase our understanding of automatic speech recognition errors. The paradigm asks human listeners to transcribe speech segments containing words that are frequently misrecognized by the system. In particular, we sought to gain information about the impact of increased con text to help humans disambiguate problematic lexical items. The long-term aim of the this research is to improve the modeling of ambiguous items so as to reduce automatic transcription errors. To this extent we have been developing a tool, the Q-ERROR graphical interface, to facilitate the analysis of automatic speech recognition errors. As previous research has shown, speech recognition errors are often modulated by a number of factors, and it can be difficult to assess the impact of each. By enabling a user to filter data in large corpora, the proposed interface can also be used to help select the relevant stimuli for human perceptual tests.
Fichier non déposé

Dates et versions

hal-01135133 , version 1 (24-03-2015)

Identifiants

  • HAL Id : hal-01135133 , version 1

Citer

Martine Adda-Decker, Ioana Vasilescu, Natalie D. Snoeren, D. Yahia, Lori Lamel. Towards Exploring Linguistic Variation in ASR Errors: Paradigm & Tool for Perceptual experiments. New tools and methods for very-large-scale phonetics research workshop (VLSP'11), Jan 2011, Philadelphie, United States. ⟨hal-01135133⟩
75 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More