Skip to content

Milestones

List view

  • Audio genre classification

    Do a simple genre classification using features from the AFE.

    Overdue by 9 year(s)
    •
    Due by May 30, 2016
    0% complete0 open0 closed

  • Speaker recognition

    Implement human speaker recognition under adverse condition (for example Toulouse recordings)

    Overdue by 9 year(s)
    •
    Due by July 28, 2016
    0% complete0 open0 closed

  • Number of sources

    Estimate the number of sound sources

    Overdue by 9 year(s)
    •
    Due by March 16, 2016
    •1/1 issues closed
    100% complete0 open1 closed

  • Localisation of synthesized sound sources

    The more advanced localisation knowledge sources like the DNN based one work not very well for synthesized sound sources. We should figure out why and try to change it in order to provide a good common localisation knowledge source that works for almost all cases.

    Overdue by 9 year(s)
    •
    Due by April 28, 2016
    •2/2 issues closed
    100% complete0 open2 closed

  • Keyword recognition

    We need keyword recognition and will integrate it in different ways, using aural and visual information.

    Overdue by 9 year(s)
    •
    Due by June 29, 2016
    0% complete0 open0 closed

  • Rule-based decision integration

    Learning of CASA grouping rules from segmented audio data. Could be based on onset and offset times, harmonicity, frequency and temporal proximity etc

    Overdue by 9 year(s)
    •
    Due by July 28, 2016
    0% complete0 open0 closed

  • Make scheduler smarter

    Complex scheduler strategy: Implement new scheduler for hybrid expert system: let scheduler follow more complex strategy to optimally identify and localize multiple sources.

    Overdue by 9 year(s)
    •
    Due by May 30, 2016
    0% complete0 open0 closed

  • Integration of identification, localisation and segmentation

    The knowledge sources for the single tasks of localisation, identification and segmentation should be integrated in order to perform an auditory scene analysis as it is for example needed by the Kopco scenario.

    Overdue by 9 year(s)
    •
    Due by April 28, 2016
    •1/1 issues closed
    100% complete0 open1 closed

  • Top-down weighting of binaural features

    Integrate with blackboard: top-down weighting of binaural features according to source models, as illustrated by modelling Kopco data. Could be tied to an attention model.

    Overdue by 9 year(s)
    •
    Due by June 29, 2016
    0% complete0 open0 closed

  • Integration with the robotic plattform

    The blackboard system is currently only used together with binaural simulations. The goal is to make it run with the actual robot as well. This will also be addressed by our Toulouse meeting in May.

    Overdue by 9 year(s)
    •
    Due by May 12, 2016
    0% complete0 open0 closed

  • Vision for other objects

    Handle the vision system for the recognition of objects in general.

    Overdue by 9 year(s)
    •
    Due by April 21, 2016
    0% complete0 open0 closed

  • Vision for human objects

    Handle the vision system for the recognition of humans.

    Overdue by 9 year(s)
    •
    Due by March 30, 2016
    0% complete0 open0 closed

  • Multi-conditional training for segmentation

    To make our segmentation stage more robust against different room conditions we should include multi-conditional training as we did it for the localisation.

    Overdue by 9 year(s)
    •
    Due by March 16, 2016
    0% complete0 open0 closed