CIESS
Constitution, indexing and enrichment of an environmental sound corpus
Corpora of environmental sounds are necessary for scientific studies of human perception and also for the creation of immersive environments (simulation, virtual reality, video games, …). This project aims to provide concrete solutions for the creation, analysis, use, and adaptation of a corpus of sounds.
Main issues and objectives
CIESS is the first project to propose a comprehensive approach to the creation, manipulation and enhancement of a corpus of environmental sounds, which extend significantly the scope of possible scientific experiments. This framework is also an essential scientific basis for the synthesis of sound urban environment optimally, implying significant industrial benefits.
From the signal processing viewpoint, the CIESS project aims to create robust and innovative methods for the automatic analysis of environmental sounds. These methods are based on the skills acquired in the community of sound recognition. The intersection of the results obtained from acoustic and perceptual ways aims at refining this sound event detection by combining acoustic and perceptual parameters. The project attempts to clarify the linkage and the interaction between these two types of parameters.
Moreover, CIESS opens the door of the modeling of urban environments. By linking a signal centered on the perceptual approach, the project will develop a coherent and meaningful modeling of soundscapes. Sector industries video game or virtual worlds, this type of modeling is very useful. From a small number of information and records, the project aims to generate almost infinite combinatorial situations.
SamoCharts
Browsing field recordings is currently mostly based on waveform and textual metadata, which may be not very informative. If the TM-chart provides an efficient tool to represent the soundscapes, it remains little used probably due to a costly human annotation. In the CIESS project, we use a new approach to compute automatically new kind of TM-charts that we call SamoCharts. We made a JavaScript implementation to compute SamoCharts based on the confidence score.
This code is available here: SAMoChart_v0.3.
Here you can download previous versions: SAMoChart_v0.2 SAMoChart_v0.1
This code allows creating a web page including SamoCharts of an audio corpus, where the confidence score of the sound events has been precomputed. In the context of the CIESS ANR project, a complete package, which includes automatic annotation algorithms, will also be downloadable.
We also build up an experiment with some files of the UrbanSound datasets. You can download the code here : UrbanSound.
This code allows creating a web page including SamoCharts of the UrbanSound audio corpus, where the levels of the sound events has been precomputed.
Collaborations
- Octogone (Project Coordinator)
- AudioGaming
People involved in SAMOVA team
Support
- ANR Corpus (2011 call)
Scheduled
- Start time : 1st february 2013
- End time : 30th january 2016