ASR performance prediction on unseen broadcast programs using convolutional neurol networks

Zied Elloumi (1, 2) , Laurent Besacier (1, 3, 2) , Olivier Galibert (4) , Juliette Kahn (5, 4) , Benjamin Lecouteux (1, 2)

1 GETALP - Groupe d’Étude en Traduction Automatique/Traitement Automatisé des Langues et de la Parole
2 LIG - Laboratoire d'Informatique de Grenoble
3 IUF - Institut universitaire de France
4 LNE - Laboratoire National de Métrologie et d'Essais [Trappes]
5 LCM - Laboratoire commun de métrologie LNE-CNAM

In this paper, we address a relatively new task: prediction of ASR performance on unseen broadcast programs. We first propose an heterogenous French corpus dedicated to this task. Two prediction approaches are compared: a state-of-the-art performance prediction based on regression (engineered features) and a new strategy based on convolutional neural networks (learnt features). We particularly focus on the combination of both textual (ASR transcription) and signal inputs. While the joint use of textual and signal features did not work for the regression baseline, the combination of inputs for CNNs leads to the best WER prediction performance. We also show that our CNN prediction remarkably predicts the WER distribution on a collection of speech recordings.

Mots clés: Convolutional Neural Networks; Performance Prediction; Large Vocabulary Continuous Speech Recognition

Domaines: Informatique et langage [cs.CL]

Zied Elloumi, Laurent Besacier, Olivier Galibert, Juliette Kahn, Benjamin Lecouteux. ASR performance prediction on unseen broadcast programs using convolutional neurol networks. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Apr 2018, Calgary, Alberta, Canada. ⟨hal-01709779⟩