I'm trying to gain some insight into acoustic voice data composed of 19 features. I want to understand what features contribute most for classification.

ADDED: Most features are related with the fundamental frequency stability. In particular I'm using voice shimmer and jitter and some related calculations.

I'm trying to use MRMR (max relevance min redundancy), but I would like to compare with some other options. ADDED: I have tried to use FeatureMiner (http://featureselection.asu.edu/index.php) which provides some interesting algorithms implementations. However many of them use deprecated Python functions and require some effort to work properly.

Are there any popular tools for these purposes?


It isn't obvious whether the 19 features are spectral and, if so, whether they contain detected transients for consonant phonics. It's not obvious whether features were extracted using RBMs (restricted Boltzmann machines), from auto-correlation in the time domain, or some other way. The intention of the classification, which would probably drive the nature of differentiation, isn't stated either. If more details are added to the question, the answers may be more detailed. Here are a few general options to compare.

  • Principal component analysis (PCA)
  • Factor analysis
  • Principal axis factoring
  • Cluster analysis
  • Regression or gradient descent if category labels exist and then checking early layer parameters for zeroes
  • Feature selection
  • Independent component analysis
  • Feature Selection: A Data Perspective, Jundong Li et. al. 2016

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.