This lecture gives an introduction into Cognitive User Interfaces (CUIs). CUIs mimic or reproduce human behavior in order to enable more natural and robust interaction between man and machine.
The lecture covers underlying algorithms, design questions, possible applications, and evaluation methods for CUIs. The focus lies on speech- and multimodal interfaces.  Training and adaptation methods for hidden-Markov-models are discussed as a main modeling paradigm for CUIs. In the domain of speech synthesis the use of Deep Neural Networks (DNN) is also discusse.
The student gets an overview of typical CUIs and their application field and should be able to understand the basic algorithms and designs.

• Introduction to CUIs
• Human cognition and perception models for CUIs
• Mixed Excitation Linear Predictive (MELP) speech coding
• Hidden-Markov-models (HMM) - Training, adaptation, recognition, synthesis
• Acoustic speech synthesis with HMM and Deep Neural Networks (DNN)
• Acoustic speech recognition
• Visual speech synthesis
• Multimodal and spoken dialog systems

Didactics:

The exercise consists of a theoretical and practical part. For the theoretical part you need to solve exercises on the blackboard. In the practical part you have to develop a speech synthesizer. For the practical part you have to work in a group of 3-5 students and record a speaker from the group and train a statistical synthesizer with the recordings and let it talk. The recordings will be made in a professional recording room at Acoustics Research Institute (ARI).

ECTS-Breakdown:
20 h Lecture + Exercise
23 h Preparation of exercise
30 h Preparation for exam
2 h written exam
------
75 h total = 3 ECTS

Participation in exercises and written exam. Each exercise counts for 20% of the grade, the exam counts for 60%. In case you cannot participate in the exam, please contact me for an individual exam (michael.pucher@oeaw.ac.at).