The study of prosody is largely absent from the rapidly-evolving field of automatic speech processing applications. Current approaches to computerized analysis of spoken text are generally heuristic and struggle for robust means of contextualization. They rely heavily on narrow semantic searches, resulting in a very partial interpretation of language and text. Specifically, technologies for automatic recognition of spoken intent still depend on vocabulary and key expressions (e.g. "ridiculous" as a dissatisfaction marker). Furthermore, they are unable to parse conversation correctly - to the point of failing to recognize where commas, question marks or periods occur. These extensively used technologies (including speech-related man-machine interfaces, speechrecognition engines and auto-translation) would be revolutionized using a computerized linguistic analysis technology that provides information regarding the context of the communication, the attitude of the speaker, emphasis and other contextual information.

The technology developed by the group of researchers requires a multidisciplinary approach, combining linguistic theory, signal processing, and artificial neural network and deep-learning methodologies. Deep machine learning algorithms for pattern recognition are combined with Saussurian pattern analysis and contextualization methodologies, for the detection of prosodic patterns and, in time, their syntactic constellations. Identification of prosodic audio features is essentially done through voice pitch, intensity, rhythm, timbre and voice quality. The features are then assigned to their contextualized "meaning" by comparison to a database, or prosodic “dictionary”. The end result can be then compressed and presented in multiple formats including xml or json.

The researchers have developed an algorithm for the detection of prosodic unit boundary, a strong emphasis detection algorithm and an analytical scheme. In addition, they are in the process of creating a "prosodic dictionary", containing a database and pattern search and currently applicable to audio files/texts. Examples of current progress are an automated classification of yes/no vs. wh- questions, advanced quantification of parentheticals, and quantification of the difference between "up-speak" and questions. Further developments are planned to make this invention applicable to written texts and to combine it with speech recognition and speech analytics, enabling a more comprehensive assessment of language and text. The invention is patent-protected.