chroma-pool
new pooling mehod designed for CNN for audio spectrogram
see chroma.py
This is a STFT spectrogram.
As shown above, horizontal lines split pixels to different areas along the frequency axis, which
are corresponding to a frequency intervals of the piano keys:
So far I do is apply chroma pool on the RAW STFT, The result is shown below:
Next to do:
Design a CNN Model
- to remove harmonics (split to different channels)
- output pure piano roll
- split details to different channel