Strengthening Human-Robot Interaction by Acoustical Implicit Communication

Participants:

Yates Yong Lin

Description:

Speaker Recognition is a technique originated in the 1970s. The main techniques used to identify speakers include

1. feature selection and extraction: Spectrum, Linear Prediction Code (LPC), Mel-Cepstrum Coefficient (mfcc), Pitch, Prosodic, Formants, Phonetics;
2. Classification Algorithms: DTW, VQ, GMM, HMM, SVM, LDA, etc.

Speaker Identification helps for the Assistive Environments in the following aspects:

1. Identify the subject´s mental state by emotion recognition-Angry, Bored, Happy, Neutral, Sad, Fear, Anxiety
2. Acoustic Human Activity Recognition-Identify the everyday activities of the subject
3. Emergency alerts about an elderly or disabled subject falling to the ground
4. Speaker Health Recognition (future research work for assistive environments)-A Pre-detection of Disease that helps remote caregivers diagnose changes in the subject's health
5. Non-linguistic Human Sound Recognition Ð to help the Deaf and the Mute (future research work for assistive environments)
   1. Promising method including glottal voice recognition, or assistive device sound recognition
   2. Other acoustical event recognition, such as TV, faucet, door, etc.

Preliminary Results

• The speaker recognition using Pitch (f0 contour) and MFCC, classification algorithms using LDA, SVM and GMM.

Publications:

1. Yong Lin, Eric Becker, Kyungseo Park, Zhengyi Le, and Fillia Makedon, “Decision Making in Assistive Environments using Multimodal Observations”, Proceedings of the 2nd International Conference on PErvasive Technologies Related to Assistive Environments (PETRA 2009), Corfu, Greece, June 9-13, 2009