Audio Applications of Matrix Completion
"Audio Applications of Matrix Completion" --- Statistics PhD student Dennis Sun
Abstract: The Short-Time Fourier Transform (STFT) is a powerful tool in the analysis and synthesis of audio. Regarding the output of the STFT as a matrix enables many audio applications, similar to techniques which have enjoyed popularity in the image processing and machine learning communities. In the first part of this talk, I describe a convex optimization approach to the classic problem of magnitude-only reconstruction --- how to recover a signal given only its magnitude spectrogram. The second part will discuss how the low-rank structure of audio spectrograms can be exploited for bandwidth extension, noise removal, and other audio application. This is joint work with Julius Smith and Rahul Mazumder.