Projects
Codage audio multipiste basé sur l'objet/
Object-based coding of multi-track audio
  In collaboration with Aquitaine Science Transfert and Laboratoire Bordelais de Recherche en Informatique

The project has been selected for presentation in Bourse aux Technologies "Publishing & Multimedia"
 
The codec in a nutshell
 
Our novel and unique technology is an object-based spatial coding system. The codec can be of use everywhere where a given sound scene is to be rendered according to the listener's preference or environment in an interactive whilst at the same time backward compatible manner. The novelty of the system lies in the generation of extra data in addition to the regular, i.e. artistic, downmix, which allows the decoder to separate the objects at the desired quality level. The extra data is generated adaptively on a block-by-block basis. This reduces the data rate significantly in comparison to residual coding. Dynamic range compression/decompression (see below) can be embedded in the codec if necessary. The codec was developed in partnership with Aquitaine Science Transfert. A C/C++ prototype is available. It features MP3 coding and decoding functionality via the LAME and mpg123 libraries.
 
Key features
 
  • Backward compatible
  • Flexible object and output format
  • Predictable sound quality
  • Perception-aware quality metric
  • Adaptive spatial resolution
  • Implicit copy protection
  • Compatible with live streaming
  • Supports mixing and mastering
  • Combinable with perceptual coding
 
Possible applications
 
  • Digital music publishing
  • Interactive music listening and DJ'ing
  • Immersive, binaural and 3D sound
  • Virtual and augmented reality
 
Teaser
 
 
Sound clips
 
Downmix stereo: wav
stereo: mp3
Extra 00 mono: wav
mono: mp3
Extra 01 mono: wav
mono: mp3
Extra 02 mono: wav
mono: mp3
Extra 03 mono: wav
mono: mp3
 
Canvas: riffs original stereo
decoded: wav
decoded: mp3
Object 00: drums original mono
decoded: wav
decoded: mp3
Object 01: bass original mono
decoded: wav
decoded: mp3
Object 02: guitar original mono
decoded: wav
decoded: mp3
Object 03: keys original mono
decoded: wav
decoded: mp3
Object 04: vocals original mono
decoded: wav
decoded: mp3
 
More examples with different quality levels can be found here.
 
Amélioration adaptative de la dynamique sonore/
Adaptive enhancement of sound dynamics
  In collaboration with Aquitaine Science Transfert and Laboratoire Bordelais de Recherche en Informatique
 
The decompressor in a nutshell
 
Our novel and unique technology is the decompressor. It is based on a mathematical breakthrough in the determination of how to invert a nonlinear dynamic system that varies over time [1]. Knowing the parameters of the compressor, it completely and accurately inverts dynamic range compression giving back the original dynamics to the squeezed signal. The decompressor has as input an audio signal and the parameters of a compressor. It uses these to generate a signal which, if it was compressed with the given parameters, would correspond to the input signal. Hence, it can be used to completely undo compression with minimal metadata. Or, it can also be used to add dynamics to a broadcast signal, regardless of whether the signal is actually compressed in the first place [2]. It requires a relatively low computational effort and has zero delay due to pure time-domain processing. So far, the technology was implemented and tested in C/C++. The screenshot below shows a graphical front end that facilitates its use. It is also available as a VST effects plugin.
 
Screen shots
 
Dynastore-X GUI

Fig. 1: Front end
 
VST Plugin

Fig. 2: Plugin
 
Video clips
 
  • Compressing and decompressing
    A demonstration of decompressor's performance. Click here.
  • Automatic adjustment
    A demonstration of the automatic mode. Click here.
 
Sound clips
 
A original
declipped
B original
declipped
C original
declipped
D original
declipped
E original
declipped
F original
declipped