This is a tool to extract instrumental track from your songs. Some code are from tsurumeso/vocal-remover, which is a deep-learning-based tool. This dummy code is intented for generating deep-learning dataset.
DISCLAIMER: ONLY INTENTED FOR GENERATING ML TRAINING DATASET, I AM NOT RESPONSIBLE FOR YOUR COPYRIGHT ISSUES, USE AT YOUR OWN RISK
It takes
- music with vocal
- music without vocal
and generates vocal
by
- Upsample the audio file for more precise alignment (with cubic interpolating)
- Calculate alignment point for beginning and end of the audio file
- Cut audio, and align the audio by resampling
- Downsampling
Advantages:
- Performs better than deep-learning-based system. (of course)
- Fixes sample alignment issues caused by software used by studios. (typically 1-50 samples for 44100HZ audio music file)
Known issues (pull requests are welcomed!):
- Can only work with original flac/wav file, otherwise may perform bad.
- Audio file must already be well-aligned at head and tail, otherwise the software will not find align points.
- Slight gliches with high-frequency voice.
- Still little BGM not filtered.
- Takes astonishing time to process