Plenary lecture 3: AI tools for speech analysis
Horia-Nicolai Teodorescu · 2010
Speech involves huge amounts of information that only recently and still partly can be analyzed with a level of sophistication comparable to the human brain. For most languages, few progresses have been done in their detailed analysis using automatic means and many linguists and phoneticians still rely on their hearing in speech evaluation. An interdisciplinary group from four institutions in Iasi, Romania, has united their efforts during more than a decade for the advancement of tools for the understanding of spoken language processes like prosody, emotional speech and personal characteristics of the voice. A consistent repository for the Romanian language, with a vast section on emotional speech was created and is available on the web. Problems overviewed in this plenary talk are the analysis and description and recognition of emotions in voice, the comparison of emotional speech characteristics in different European languages, interaction of grammar and speech in specific grammatical constructions, spoken language statistics of the Romanian language, and pathologies' effects on speech. We overview the principles of the computational tools developed and the results of speech analysis, with an emphasis on spoken language statistics and emotional analysis and recognition in speech. The tools presented include GRID applications for statistical analysis of voice, serial programs for automatic speech pattern identification with biometric and data mining applications, programs for precise determination of the prosodic traits like voice pitch, and programs for statistical characterization of emotional speech.