Skip to content

Latest commit

 

History

History
12 lines (11 loc) · 439 Bytes

TODO.md

File metadata and controls

12 lines (11 loc) · 439 Bytes

TODO

  • Accuracy instead of lexical error rate
  • Decoding real-time audio data
  • Try using some kind of feature extraction. Real-valued DFT looks promising, since it can radically decrease the amount of training data (the signal is very narrow band, only a few DFT buckets contain most of the energy).
  • Train a network to identify multiple signals in audio data and mark them with their approximate frequencies.
  • ???
  • Profit.