Abstract
Corpus-based concatenative synthesis (CBCS) (Schwarz, 2007) utilises segmentation and analysis of audio sources as source material to create a synthesised audio stream. This is achieved by creating a corpus containing multiple audio segments, analysing each segment individually, performing matching of segments to target values based on this analysis, and finally concatenating the results. Which source audio file a segment comes from, or its chronological position within this source file in relation to other segments, is typically discarded. Many playback synthesis approaches can be found within the field of CBCS covering a wide range of sound, from more granular approaches focused on the moment-to-moment sound, to more rhythmic approaches, and even ones focused on the larger overall structure of sound (Zils, 2001).While some tools (e.g. AudioGuide) do not feature any visualisation, most existing CBCS
software (e.g. CataRT and AudioStellar) presents a two-dimensional representation of the corpus as a point cloud where each segment is assigned a position in the space based on selected descriptor values that are converted to X and Y co-ordinates for display. I refer to such tools as visualised corpus browsers (VCB). However, not all VCBs make use of CBCS (e.g. XO), as these visualisations can also be used for organisation of large sample libraries (Zils, 2001). Visualisation in VCBs in a three-dimensional space is still a largely unexplored area.
In this research, I aim to explore both the three-dimensional visualisation potential of VCBs, as well as to experiment with novel approaches to CBCS, particularly through the use of chronological segment information within the synthesis engine. Created in response to these aims, I present Audio Corpus Explorer (ACorEx), an open-source software for three-dimensional visualisation and interaction with an audio corpus that takes a new approach to corpus-based concatenative synthesis, addressing the second of these aims.
This commentary starts with an introduction to the concepts of CBCS and VCBs in Chapter 1. This is followed by explanations of the ideas of corpora, audio descriptor analysis, and dimensionality reduction, as well as a review of existing CBCS and VCB tools in Chapter 2. Chapter 3 briefly discusses the evolution of my research methodology, before a detailed description of the development process and outcomes of ACorEx is given in Chapter 4. This is followed by a post-mortem discussing ACorEx and its potential for musicality and three-dimensional navigation and visualisation in Chapter 5. Finally, Chapter 6 outlines my conclusions, and discusses further research possibilities within this space.
| Date of Award | 12 Jun 2025 |
|---|---|
| Original language | English |
| Supervisor | Alex Harker (Main Supervisor) |