Skip to content
A deep-learned time-stretch algorithm for audio
Python
Branch: master
Clone or download

Latest commit

Fetching latest commit…
Cannot retrieve the latest commit at this time.

Files

Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.gitignore
LICENSE
README.md
export_audio.py
load_data.py
lossplot.png
main.py
models.py
parameters.py

README.md

Super Resolution Audio Stretch (SRAS)

A deep-learned time-stretch algorithm for audio

Training data consists of MIDI files rendered by a string emulation synthesizer. The output audio files for training are created by playing the MIDI file at half the BPM as the input. Thus, the input audio files are half the length of the output files.

I am currently experimenting with different architectures for the neural net and will update this README when I've achieved better results.

TO-DO:

- Spectrogram conversion/data handling

- Spectrogram pre-upscale model

  • Spectrogram post-upscale model
  • SR-GAN using audio spectrograms
You can’t perform that action at this time.