Per-Tone Equalization and Per-Tone Precoding for Speech and Audio Signal Processing

Per-tone equalization and per-tone precoding have been developed as suitable tools for digital subscriber line (DSL) transmission in highspeed digital modems. DSL adopts so-called discrete multi-tone transmission (DMT) as a modulation format, which is essentially a Fourier transform based transmultiplexing operation that has indeed been shown to allow for the additional per-tone processing. In this project per-tone equalization and per-tone precoding will be developed as suitable tools i) in a filter bank subband processing setup instead of a transmultiplexer setup, and ii) for a very different signal processing application field, namely speech and audio signal processing. With this, the current toolset of transform domain (e.g. frequency domain) and filter bank subband processing is extended, and performance improvement in specific applications can be sought for. Fundamental aspects will be studied, related to perfect reconstruction theory and aliasing, as well as as signal adapted weighted-overlap-add and filter bank design. Two specific speech and audio signal processing cases will be investigated, namely pertone equalization based multi-channel noise reduction and per-tone precoding based sound zoning.