/DFSMN-Based-Lightweight-Speech-Enhancement

Deep Feedforward sequential memory networks(FSMN)

Primary LanguagePythonApache License 2.0Apache-2.0

DFSMN based light weight speech enhancement model

under construction

To do

  • use rezero to control skip-connection
  • real spec predict cirm
  • clp predict cirm
  • deep filter
  • online/dynamic augmentation
  • ncnn inference for android/linux
  • windows desktop app