A hierarchical approach to obstruent phoneme recognition in continuous speech is presented. This approach is based on the division of the recognition process into smaller tasks which are easier to design, develop and optimise. The developed system first distinguishes between obstruents and other phonemes and then proceeds to classify the phonemes. Performance of around 80% is achieved. The system's tolerance to temporal misalignment makes it a good candidate for recognition in fluent speech.
Keywords: Obstruent Phonemes, Hierarchical, TDNN