Winter 2020/2021 Theme: Signal Processing with Machine Learning
Signal processing by optimal filtering builds on rigorous models of the distorted signals, e.g., linear models or additive (white) Gaussian noise. Implementation then, however, requires precise specification of a-priori information over the model, e.g., in terms of target-signal and observation-noise covariances and/or spatial steering vectors, which is frequently not available and thus invokes heuristic workaround. Real systems in the field acoustic signal processing and digital communication, moroever, exhibit additional characteristics of nonlinearity and/or timevariance, which are hard to capture with closed-form models.
The seminar will thus introduce the frameworks of aritificial intelligence, machine learning, and adaptive systems. Those disciplinces offer great support for model compensation by utilization of data on versatile architectures of computational networks. Students and instructors will follow voice-and-video online lecture series, including programming assignment and quiz, to jointly and interactively investigate the link between machine learning and signal processing. Moderation of the seminar will be shared among students and instructors regarding fundamentals of machine learning and specific applications of audio and acoustic signal processing:
• Machine Learning Support for Speech Enhancement
• Data Fusion over Acoustic Sensor Networks
• Signal Processing for Virtual Acoustic Reality
- Kursleiter/in: Aleksej Chinaev
- Kursleiter/in: Gerald Enzner