Kolloquium - Details zum Vortrag
Sie werden über Vorträge rechtzeitig per E-Mail informiert werden, wenn Sie den Newsletter des kommunikationstechnischen Kolloquiums abonnieren.
Alle Interessierten sind herzlich eingeladen, eine Anmeldung ist nicht erforderlich.
Master-Vortrag: Investigations on Joint Application of Blind Source Separation and Primary-Ambient Extraction to Stereo Signals
Si Hu
Dienstag, 23. April 2019
14:00 Uhr
Hörsaal 4G
Due to the rapid development of surround sound systems, audio files in multi-channel format are increasingly required. In consideration of the fact that most legacy audio recordings are in stereo format, it is favorable to convert these audio recordings in stereo to multi-channel format which refers to upmixing technology. In the upmixing process, so-called primary and ambient components play an important role. They are extracted from the input stereo signal by means of Primary-Ambient Extraction (PAE). There are many PAE approaches available, but they have a common shortcoming. When they handle audio signals with multiple concurrent primary sources, the extraction quality will drastically degrade and the listening experience will be significantly impaired.
In order to address this issue, we consider to incorporate Blind Source Separation (BSS) into the PAE process. In BSS, it is investigated how to recover every single sound source from a mixed audio signal. The basic idea is that an audio signal with multiple concurrent primary sources can be separated into several sub-signals by means of BSS. Each sub-signal is expected to contain only one primary source. Then PAE is executed for every single sub-signal. Theoretically, PAE processing should be facilitated, because PAE works well when dealing with the signals with only a single primary source.
In this thesis, the feasibility of this idea is validated. Non-negative Tensor Factorization (NTF), as a prevalent BSS technique, is employed in the experiments. We propose several NTF based strategies to execute PAE on audio input signals with three primary sources. A series of numerical experimental results and subjective listening tests show that the PAE performance can be enhanced in this way in the case of multiple primary sources.
