Notebooks - michele-perrone/SpectrogramPlayer GitHub Wiki

Spectrogram_Player.ipynb

This is the main notebook. Here, we provide the code to perform a comparison of different existing approaches for the phase reconstruction of magnitude-only audio spectrograms.

This notebook can be used to download the datasets and store them in separate folders located on Google Drive. This is done to avoid downloading the datasets every time we need them in one of our notebooks.

GstPEAQ_testing.ipynb

In this notebook, we use GstPEAQ to test the results obtained by the various techniques.
The audio clips must be available as .wav audio files.
Please note that GstPEAQ comes only as Gst plugin or as a command-line utility. We will use it as the former by creating command strings to be executed on the local terminal. Therefore, you must install GstPEAQ on you local machine in order to run this notebook.

Here, is the link to the GstPEAQ Repo that we used.

NeMo_Spectrogram_Testing.ipynb

This is a testing notebook. Here, we try to understand what are the exact arguments, outputs and parameters of the following methods:

- SpectrogramGenerator.generate_spectrogram(tokens=parsed)
- Vocoder.convert_spectrogram_to_audio(spec=spectrogram)

The goal is to use them independently, without running the entire end-to-end model.

Numpy_to_WAV.ipynb

This notebook is used for converting numpy arrays containing audio fragments into wav files. The audio fragments are both the originals and the reconstructed ones. The WAV files are then used to compute the PEAQ metrics.