Modeling pronunciation variation for ASR: A survey of the literature

Non-Native Pronunciation Variation Modeling for Automatic Speech Recognition

“…Datadriven methods can be further classified into direct data-driven or indirect data-driven approaches. The direct data-driven approach derives pronunciation variants depending on pronunciation training databases, as proposed in Fosler-Lussier, 1999;Strik et al, 1999). When an ASR system employs the adapted pronunciation dictionary using a direct data-driven approach, some unseen words might appear during ASR testing.…”

“…Usually, the variant pronunciations for each word are added to the pronunciation models, which is similar to a multiple pronunciation dictionary approach Fosler-Lussier, 1999;Goronzy et al, 2004;Gruhn et al, 2004;Raux, 2004;Strik et al, 1999). The pronunciation variants from non-native speakers can be derived by either knowledge-based or data-driven approaches (Strik et al, 1999).…”

“…Usually, the variant pronunciations for each word are added to the pronunciation models, which is similar to a multiple pronunciation dictionary approach Fosler-Lussier, 1999;Goronzy et al, 2004;Gruhn et al, 2004;Raux, 2004;Strik et al, 1999). The pronunciation variants from non-native speakers can be derived by either knowledge-based or data-driven approaches (Strik et al, 1999). Note that knowledgebased approaches are based on linguistics or phonetic knowledge (Schaden, 2003;Tajchman et al, 1995;Wiseman et al, 1998), whereas data-driven approaches automatically derive pronunciation variants from non-native speech data and can be further classified into either a direct method Fosler-Lussier, 1999;Strik et al, 1999) or an indirect method Fosler-Lussier, 1999;Goronzy et al, 2004;Svendsen, 2004;Wolff et al, 2001).…”

“…There are two approaches pertaining to pronunciation model adaptations for non-native speech: a knowledge-based approach and a data-driven approach (Strik et al, 1999). A knowledge-based approach uses pronunciation rules from phonological knowledge and develops a pronunciation dictionary based on the pronunciation rules.…”

Towards capturing fine phonetic variation in speech using articulatory features

“…Furthermore, the modelling of spontaneous speech is a difficult issue for phone-based ASR systems. Many techniques and approaches have been tried to model spontaneous speech phenomena such as coarticulation, but only to limited successes (for an overview, see Strik & Cucchiarini, 1999).…”