Human beings extract a great amount of information from the speech we hear, both consciously and unconsciously. Besides understanding what is being said, we know who is speaking (when the voice sounds familiar), what language they speak (if it is one of the languages we know, even superficially); we can even estimate whether the speaker is angry, happy, or tired; and we perceive things as subtle as irony and sarcasm.
One of our research directions at LIAA is the automatic extraction of these kinds of information from the speech signal. We work, among other tasks, on the recognition of the speaker’s identity, language, emotion and trust level. We are also working on a system for detecting pronunciation mistakes in second language learners, aiming at developing a virtual tutor for language learning.