Using Machine Learning for Speech Extraction and Translation: HiTEK Languages

Speech processing deals with retrieving and vocalizing conversational words/sentences i.e. articulatory phonetics, manner of articulation, place of articulation, articulatory gestures, articulatory phonology, articulatory speech recognition, and articulatory synthesis from Multilingual Source to Target language. The research focuses on multilingual speech recording in a single utterance and translation to a target language. Greedy method is used for fetching speech from the user. It consists of the grammatical structures of the speech in the dictionary using cohesion based method for term similarity. It translates speech to text then maps text to a set of phones resulting in target language speech. Multilingual Supervised Speech Dictionary is built for speech to speech translation, currently consisting of four languages Hindi, Telugu, English and Kannada with 100 phones for each language.