Kolloquiumsvortrag (ET&IT), M.Sc. Jonas Sauter, Nuance Communications / am 20.11.2017

20.11.2017 von 17:15 bis 18:00

Institute Ostufer, Geb. D, "Aquarium", Kaiserstr. 2, 24143 Kiel

Titel: Artificial Bandwidth Extension for Speech Signals Using Deep Neural Networks

Abstract: In mobile communication, the bandwidth of transferred speech signals is either narrow-band (300Hz – 3.4kHz) or wide-band (50Hz – 7kHz or higher). As the limitation to 3.4kHz degrades the speech quality and intelligibility, it is of great interest to artificially extend narrow-band speech signals to wide-band speech.

This talk presents a deep neural network (DNN) approach to artificial bandwidth extension with a focus on robustness in practical applications.

It is based on the source-filter model which decomposes the signal into two parts: an excitation signal and a spectral envelope. The excitation (source part) describes the fine spectral structure which consists of white noise for unvoiced speech and an impulse train for voiced speech. The spectral envelope (filter part) describes the coarse spectral structure, i.e. the formants or resonance frequencies that make up different phonemes.

While the extension of the excitation signal can be done with simple mathematical methods that do not introduce strong artifacts, the envelope is much more relevant for the quality of the reconstructed wide-band signal. That is why the wide-band envelope is estimated with DNNs in this approach, which are trained on a large speech corpus.

Short biography

Jonas Sautter studied Electrical Engineering, Information Technology and Computer Engineering at RWTH Aachen University, Germany. He received his Master of Science degree in 2016. The Master’s thesis with the title “Digital Robust Control for Active Noise Cancellation in Headphones and Hearing Aids” was composed at the Institute of Communication Systems at RWTH Aachen. Since November 2016, he is a PhD student at Nuance Communications in Ulm, supervised by Professor Gerhard Schmidt, Head of the Digital Signal Processing and System Theory group at Christian-Albrechts-Universität, Kiel.

Prof. Schmidt

Diesen Termin meinem iCal-Kalender hinzufügen