Sound synthesis and sound processing for audio and musical applications

该研究项目旨在通过利用信号处理技术的最新进展来提高音频信号表示和建模的准确性和充分性。更准确地说，目标是更好地解决声音信号的双重性质，声音信号通常是时间瞬变和调制频率分量的混合物。它将导致更好的声音处理质量，声音合成，并将培育应用项目，如高质量的声源分离或音频变形。 这个项目是所谓的声音合成和声音处理的分析/合成方法的一部分，这是一种在音乐应用中经常使用的方法，也是声音设计，音频编码和录音的一部分。分析/合成的基本目标是构思相关的模型，以通过一组控制合成结构的时间函数（控制参数）来表示声学信号。这些时间函数可以抽象地生成，或者在分析过程中从预先录制的声音中提取。重新合成产生一个信号，听起来与原始声音相同。通过修改和替换控制参数，该分析-合成方案可以提供对声音材料的非常精细和精确的处理。例如，它可以允许从单个原始声音信号产生一系列合成声音信号，或者在两个关键声音之间进行变形。它也可以用于通过将时间函数适当排序和分组为相干子集来对源分离进行决策。在这种情况下，音频信号的表示质量是至关重要的。 拟议的研究计划是双重的：在基本层面上，研究工作包括加强声音信号的时频表示和参数建模的核心基础。在实践层面上，研究的重点是音频应用，可以受益于这些新的信号处理工具。

This research project aims at improving the accuracy and the adequacy of audio signal representations and modelling by benefiting from recent advances in signal processing techniques. More precisely the goal is to better address the dual nature of sound signals, which are often a mixture of temporal transients, and modulated frequency components. It will result in a better quality of sound processing, sound synthesis, and will nurture applied projects such as high quality sound source separation or audio morphing. This project is part and parcel of the so-called analysis/synthesis approach for sound synthesis and sound processing, an often used approach in musical applications, but also in sound design, audio coding and sound recording. The basic goal of analysis/synthesis is to conceive relevant models to represent acoustical signals by a set of temporal functions (the control parameters) that controls a synthesis structure. These temporal functions can be generated in abstracto or extracted from a pre-recorded sound during the analysis process. Re-synthesis produces a signal that sounds perceptually identical to the original sound. By modifying and substituting control parameters, this analysis-synthesis scheme can provide for very refined and precise processing of sound material. For example, it may allow one to produce a family of synthetic sound signals derived from a single original one or to carry out a morphing between two key sounds. It can also be used to make decision on source separation by a proper sorting and grouping of temporal functions into coherent subsets. In this context the quality of the representation of audio signals is essential. The proposed research program is two-fold: at a fundamental level the research work consists in strengthening the core foundation of time-frequency representation and parametric modelling of sound signals. At a practical level the research is focused on the audio applications which could benefit from these new signal processing tools.