Speaker: Vikram Mitra (Apple, Inc.)
Title: "Speech as a behavioral signal for mental and cognitive health: Some observations and results"
Abstract: Speech is a key signal for human-to-human communication and for human-to-machine communication. It is used to convey our happiness-anger, satisfaction-frustration, excitement-despair and many other emotions. Speech also acts as a behavioral signal for humans to share their state-of-mind. Recognition of speech, speaker and language has been the center stage for speech technology research and development, and with recent advances, the state-of-the-art performance has progressed to admirable heights. Existing speech technologies are capable of deciphering what has been said with a high accuracy but often fails to decipher how it has been said, limiting the capabilities to detect human behaviors such as satire, frustration, negative thoughts, forgetfulness, etc. Human listeners are capable of comprehending both the meaning and the intent of speech with exceptional accuracies; however, our existing speech processing systems are limited only to the meaning, often failing to comprehend intent. Recent studies have investigated the use of speech to detect a speaker’s emotion, mental and cognitive health. This talk will explore speech as a behavioral signal to uncover the speaker's state of mind, particularly their mental health. Psychomotor retardation is a behavioral manifestation of mental health condition, which is an outcome of reduced articulatory motions. This talk will investigate if speech articulatory patterns can be used to uncover the speaker’s psychomotor retardation and hence be used as a behavioral signal to detect mental health conditions. In addition, the talk will present how robust speech representations and articulatory information can be used to improve acoustic modeling tasks for speakers with certain clinical conditions.