Phoneme-based Recognition of Finnish Words with Dynamic Dictionaries

Abstract

In this paper we present an isolated -word recognition system using first HMM to recognize the underlying sequence of phonemes, then DP and phoneme n -gram matching techniques to determi ne the corresponding nearest idealized phoneme sequences in the dictionary. Our approach is based on the observation that there is almost a one -to-one match between phonemes and letters in the official written Finnish language and a representation of the words in the dictionary as phoneme sequences. A significant advantage of our system lies in its ability to easily modify the dictionary without retraining the ASR models. Speaker dependent word recognition experiments can show an achievement of 95% recognition rates.

2 Figures and Tables

Cite this paper

@inproceedings{Seydou2001PhonemebasedRO, title={Phoneme-based Recognition of Finnish Words with Dynamic Dictionaries}, author={F. Seydou and Tapio Sepp{\"a}nen and Johannes Peltola and Pekka V{\"a}yrynen}, year={2001} }