ICASSP 2020 | Paper ID: 5769

Addendum (Updated April 12, 2020)

The original paper (accepted ICASSP 2020) is here

Paper Title

IMPROVED SPEAKER INDEPENDENT DYSARTHRIA INTELLIGIBILITY CLASSIFICATION USING DEEPSPEECH POSTERIORS

[A] Typo

Then, posterior probabilities from all the classifiers for each class are multiplied and the class with maximum probability is chosen as the classification of ξ, namely,

should be correctly read as ..

Then, posterior probabilities from all the classifiers for each class are multiplied and the class with maximum probability is chosen as the classification of ξ, namely,

[B] Additional Results

Note:

  • Features
    • IS09 (I) (384): See Table 3 here
    • IS10 (1583): See paper
    • DsL (D) (2100): Derived from DeepSpeech ASR (New)
    • QDsL (Q) (11): Non-linear quantization of DsL
  • FC: Feature Combination
  • PF: Probability Fusion