Multilingually trained bottleneck features in spoken language recognition. (November 2017)

• Record Type: Journal Article
• Authors: Fér, Radek; Matějka, Pavel; Grézl, František; Plchot, Oldřich; Veselý, Karel; Černocký, Jan Honza
• Is Part Of: Computer speech & language. Volume 46(2017)
• Page Start: 252
• Page End: 267
• Publication Date: 2017-11
• Subjects: Multilingual training -- Bottleneck features -- Spoken language recognition; Speech processing systems -- Periodicals; Automatic speech recognition -- Periodicals; Computers -- Periodicals; Linguistics -- Periodicals; Speech-Language Pathology -- Periodicals; Traitement automatique de la parole -- Périodiques; Reconnaissance automatique de la parole -- Périodiques; Automatic speech recognition; Speech processing systems; Electronic journals; Periodicals; 006.454
• DOI: 10.1016/j.csl.2017.06.008
• Languages: English
• ISSNs: 0885-2308

Abstract

Highlights: Practical aspects of multilingual training of bottleneck features for spoken language recognition are shown. Multilingually trained bottleneck features are demonstrated to be a better choice for spoken language recognition than bottleneck features trained on just single language. This is further demonstrated by showing the insensitivity to target language in i-vectors space. Different configurations of bottleneck neural networks are evaluated from language recognition point of view but also by looking at speech recognition performance of resulting bottleneck features. Fusion experiments demonstrate the complementarity of both types of features (mono- and multi-lingual). Abstract: Multilingual training of neural networks has proven to be simple yet effective way to deal with multilingual training corpora. It allows to use several resources to jointly train a language independent representation of features, which can be encoded into low-dimensional feature set by embedding narrow bottleneck layer to the network. In this paper, we analyze such features on the task of spoken language recognition (SLR), focusing on practical aspects of training bottleneck networks and analyzing their integration in SLR. By comparing properties of mono and multilingual features we show the suitability of multilingual training for SLR. The state-of-the-art performance of these features is demonstrated on the NIST LRE09 database.