Phonemic Acoustic Modeling

Abstract

As previously discussed, one of the major problems in automatic speech recognition (ASR) for dialectal Arabic is the sparse speech resources and the limited research done in phonetic transcription. In this chapter, we address the problem of having only a small amount of dialectal Arabic speech data, for which it is possible to have a phonetic transcription. Initially, a phonemic acoustic model was trained using a large amount of Modern Standard Arabic (MSA) speech data. After applying phoneme sets normalization, cross-lingual model adaptation was performed using dialectal speech data. The proposed approaches in this chapter were applied with Egyptian Colloquial Arabic (ECA).