Open yusufkhanmohammad opened 4 years ago
Mohammad, I suggest you try to make a spectrogram from just the first file in the flac directory. This first call we term S 1 and is the most common call that the local orca use. It should give you a spectrogram similar to the one below I just made with Audacity from that first file. Val [image: image.png]
On Wed, Mar 18, 2020 at 12:00 PM Mohammad Yusuf Khan < notifications@github.com> wrote:
spectrogram.ipynb uses soundfile library and makes a spectrogram a bit hazy, tried doing with Librosa amplitude_to_db spectrogram AtoD = librosa.amplitude_to_db(np.abs(librosa.stft(data)), ref=np.max)
https://colab.research.google.com/drive/1LRdUcsKTiBI-ERVJQ0OEImcUo7WqtjrN
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/orcasound/orcadata/issues/27, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABFXKXMG5UOTNWUE3ZWE3JDRIEK3LANCNFSM4LOYLI3A .
Hey, a good way to start enhancing the spectrogram would be to remove noise from the audio signal before computing the spectrogram (maybe using a package like open-unmix-pytorch). Some possible ways to enhance the spectrogram are:
More information can be found in this interesting research paper: http://aircconline.com/sipij/V3N2/3212sipij01.pdf
Interesting paper @atreyamaj Here is the first stage of preprocessing that I performed on S_1 .wav file. I think this notebook will help to get a basic look of preprocessing https://colab.research.google.com/drive/1ckW8EaIO9Vf1n4KZHz289aryWKrclHtF
This notebook is great! Thanks for sharing it @kunakl07 , I'm sure it'll help others looking to preprocess data too!
Welcome @atreyamaj , the actual credits go to Abhishek Singh and Jesse Lopez who performed these preprocessing tasks. I have used it to apply on S_1.wav file which is here.
@yusufkhanmohammad Hey, I just came across a python library that can be used for denoising. The detailed steps for how it does so are given in the documentation. Link: https://pypi.org/project/noisereduce/
spectrogram.ipynb uses soundfile library and makes a spectrogram a bit hazy, tried doing with Librosa amplitude_to_db spectrogram
AtoD = librosa.amplitude_to_db(np.abs(librosa.stft(data)), ref=np.max)
https://colab.research.google.com/drive/1LRdUcsKTiBI-ERVJQ0OEImcUo7WqtjrN