ICSI Logo    EARS: Novel Approaches Tasks

Signal processing - Replacing the spectral envelope by long-time and short-time (multirate) probabilistic functions of the spectro-temporal plane.

The spectral envelope is a fragile information carrier. We propose to use probabilities from multiple time-frequency patches. There are several areas of study for this approach:

  • Incorporating temporal trajectories of narrow subbands
  • Optimal search for more general patches
  • Alternative functions to represent these patches (e.g., cepstral or autoregressive)
  • Generating data-driven broad class probabilities to serve as the features

    Statistical Modeling - Modifying the statistical models, both to incorporate these new multirate front ends and to explicitly handle areas of missing information. Modeling features that span different time scales and dependence across scales/streams.

    Why multirate vs. redundant features?

  • Redundant features violate conditional independence assumptions, and can lead to poor confidence (posterior) estimates
  • Redundancy adds unnecessary computation

  • Important research issues:

  • Acoustically driven rate mixing and/or variable alignment
  • Discriminative learning of dependence across streams
  • Back to the EARS main page