Vowel-Consonant Speech Segmentation by Neuromorphic Units

For the time being speech is still a much complex process far from being fully understood. To gain some insight on specific open problems in its automatic treatment (recognition, synthesis, diarization, segmentation, etc.) neuromorphisms and knowledge derived from the understanding on how the Auditory System proceeds may be of crucial importance. The present paper must be seen as in a series of preliminary work carried out trying to translate some of this understanding to solve specific tasks as speech segmentation and labelling in a parallel way to the neural resources found in the Auditory Pathways and Cortex. The bio-inspired (neuromorphic) design of some elementary units covering simple tasks as formant tracking or formant dynamics is exposed. In a further step it is shown how simply neural circuits employing these units may convey successful vowel-consonant separation independently of the speaker. The paper is completed with the discussion on how this processing may be used to develop specific applications as in Speech Segmentation and Diarization and in Speaker Characterization.