Skip to content

Latest commit

 

History

History
45 lines (34 loc) · 2.07 KB

Considerations.md

File metadata and controls

45 lines (34 loc) · 2.07 KB

Considerations

There are a lot of choises using a library, framework so on and so forth. Here are some of my considerations, which may be wrong or outdated.

Datasets

  • DeepShip (22gb, 4 classes)
  • ShipEar (awaiting replay)

Audio processing libaries

During this project i have used nnAudio and torchaudio. I prefare torchaudio because it handles the io, supports read offsets as argument, and even normaizes audio by default. with that said nnAudio is also a great, and is a powerfull library.

Signalprocessing

Mel Spectrograms

In audio classification with neural networks it is common to use mel spectrograms.
Mel are designed for speech. That may not be ideal for ship classification. For instance, in speech, there are no useful spectral information below our pitch. As a result the firtst mel filterbank "removes" that information. That is not ideal for classifying ships that can radiate noise at those frequencies.

STFTs

With Short time Fourier transform (STFTs) you need to know what kind of temporal or spectral resolution you need.

CQTs

Constant Q Transforms (QCTs) are more flexible than FTFTs. The balance time/ spectral resolution, but are heavier to compute.

DNN Library

fastai & pytorch was for me a littlebit faster when training models compared to tensorflow, even with almost identical model.