About testing noisy wav files

[zeynepgulhanuslu]

Hi, Thank you for sharing this code. I've two question:

1. When I run the code with audio files on Kaggle challenge dataset, it doesn't decode the last morse code. When I examine the code, I find "..-" decoded for F, but the last dot not decoded. What could be the reason for this?

example : FEFOFQUHYNZ3WONZRETF output : FEFOFQUHYNZ3WONZRET

2. What would you suggest for decoding noisy wav files with negative snr values? Is it helpful to change parameters? Which parameters should I change?

Thank you Have a nice day

[ag1le]

Hi zeynepgulhanuslu

1. The likely reason is that there is not long enough pause after the last dot, or that there is no transition after that pause. I haven't looked at this code for many years ...

2. The largest improvement would come from filtering out the noise. You could try different values for the cut-off frequency - instead of 40 Hz you could try 25 Hz and see if that improves decoding accuracy. See the line below that is responsible for that: Wn = 40./ (Fs/2.) # 25 Hz cut-off for lowpass

[zeynepgulhanuslu]

Thank you i will try to do that. Do you think your LSTM morse decoder will give better results for noisy dataset?