My participation log of the Bandai Namco Data Science Challenge competition

This repo contains my solution to the Bandai Namco Data Science Challenge competition.

Overview

The objective of this competition was to remove the artificial noise from the given voice data. The voice data was converted into Mel Spectrogram.

Dataset

• raw[0-9]\{3}.npy
  • Mel Spectrogram of raw voice data
  • 100 files
• noised_tgt_[0-9]{3}.npy
  • Mel Spectrogram of noised voice data
  • 30 files

My solution

I used a pix2pix model to achieve this competition.

Prerequisites

• become-yukarin
• librosa
  • for convert_to_wav.py
• numpy
  • for convert_to_wav.py

License

MIT License