ABSTRACT
Pronunciation teaching methods, as a part of computer assisted language learning systems, are currently limited in their ability to produce feedback on pronunciation quality. After an overview of previous work on pronunciation teaching, this article presents a pronunciation scoring algorithm based on automatic speech recognition, whereby scores at a phonemic level can be calculated. These ‘goodness of pronunciation’ scores consist of a likelihood ratio between forced alignments and a maximum likelihood monophone loop. The results of evaluation experiments demonstrate the method’s capability of detecting both individual mispronunciations, and they give a general assessment of which sounds a student tends to pronounce badly.
