We consider a spectrogram of a music excerpt, e.g.:
Playing-input.mp4
The aim of this code is to split it into three components:
- the sinusoidal component,
- the transients component, and
- the noise component.
The results are the folowing:
Playing-sinusoid.mp4
Playing-transient.mp4
Playing-noise.mp4
We synthesize according to the recovered lines:
The code is organised in the following way:
core/contains the source code.data/contains the data used in the experiments. The subfolderaudio/contains the input audio files and the subfolderobjects/constains the STFT layers with different time and frequency resolutions.procedures/contains the procedures used to apply the morphological pipeline, synthesize and plot.results/contains the results obtained in the experiments.testscontains some tests to check the code.main.pyis the main script that should be run to execute the pipeline and plot the outputs. The input audio is encoded in the variablename.settings.pyis a file with the settings used for each audio sample for running the code.
If you want to change the parameters used for the computations, see the file core/parameters.py.
If you want to ear an audio sample, go to results and find the name of the sample. Inside its folder, there are the arrays, the audio and the images. In the audio folder there are six different files:
input.wavis the original audio sample.sinusoids.wavis the output of the sinusoidal component.transients.wavis the output of the transient component.filtered_noise.wavis the output of the noise component.denoised.wavis the sum of sinusoids and transient.output.wavis the sum of sinusoids, transient and noise.