The main.py file is for training a single SIREN model, conducting a full sweep, or fine-tuning a previous model.
The main_stitching.py file is for stitching a 10 second sound sample together by compressing 1-second segments at a time and then stitching them together at the end. This can be done by either training each from scratch or using a warm-start method.
The testing.ipynb and util.py documents the implementation of quantization, and the general testing aswell as the metric calculations conducted in the report.
During this project, sounds clips have been used for testing, consisting of a Classical, Rock, Pop and Speech sample. All can be found in the soundFiles folder, along with their corresponding MP3 and Opus formats.
All trained models used during the report can be found in the state_dicts folder.
This work is using ViSQOL as a metric. Installation guide can be found here: https://github.com/google/visqol