Phoneme transcription based on Sampa for Norwegian

Abstract

This paper describes a backpropagation neural network that learns to convert Norwegian text to phonemes. The corpus consists of about 50000 Norwegian words and their transcriptions. The corpus is developed by the Norwegian Telecom research centre. The transcription scheme used is Sampa for Norwegian. The performance of the system has been tested on about 10000 unknown Norwegian words. All the experiments have been done on a PC.