Interspeech'2005 - Eurospeech
Speaker independent name speech recognition has become hot application in handheld devices such as mobile phones and personal digit assistants (PDAs). This paper presents a new mixed-lingual ASR system that will enable Chinese mobile phone users to conduct Mandarin and English name speech recognition simultaneously without switching language modes. We created an elaborately designed mixed acoustic model set which consists of two monolingual acoustic model sets, i.e. context-dependent Mandarin models and context-independent English models. Design of such mixed acoustic models for mobile phones is to balance the complexity and performance of the ASR system. We also described Mandarin and English mixed-lingual letter-to-sound converter in this paper. We demonstrated the effectiveness of the mixed-lingual name speech recognition system through off-line ASR experiments. Furthermore, the mixed-lingual name ASR system has also been integrated in Motorola new PDA phones and showed good performance in real world application.
Bibliographic reference. Ren, Xiaolin / He, Xin / Zhang, Yaxin (2005): "Mandarin/English mixed-lingual name recognition for mobile phone", In INTERSPEECH-2005, 3373-3376.