D-SPEECH

WIP. speech processing (especially, speech recognition) toolkit for D

feature

• minimal
• efficient
• faster

roadmap

• dsp.feature module
• dsp.learn.acoustic module
• dsp.wfst module

structure

• source/dsp/

  • feature: speech feature (FBANK, MFCC) extraction module
  • learn: machine learning module
    • acoustic: acoustic model module
      • hmm: hidden Markov model based sequence transducer
      • gmm: Gaussian mixture model based acoustic model module
      • neural: nueral network based acoustic model module
    • language: language model module
      • ngram: n-gram based language model module
      • neural: neural network based language model module
  • wfst: weighted finite state transducer module
    • container: sequence acceptor and transducer containers
    • algorithm: wfst related algorithms

• bin/

  • dsp-feature: speech feature (FBANK, MFCC) extractor command