Datasets:
metadata
dataset_info:
features:
- name: image
dtype: image
- name: audio_file
dtype: string
- name: slice
dtype: int16
splits:
- name: train
num_bytes: 1903861364.293
num_examples: 10663
download_size: 1903696036
dataset_size: 1903861364.293
pretty_name: Mel spectrograms of music
size_categories:
- 10K<n<100K
source_datasets: []
tags:
- audio
- spectrograms
task_categories:
- image-to-image
task_ids: []
Over 20,000 512x512 mel spectrograms of 5 second samples of music from my Spotify liked playlist. The code to convert from audio to spectrogram and vice versa can be found in https://github.com/teticio/audio-diffusion along with scripts to train and run inference using De-noising Diffusion Probabilistic Models.
x_res = 512
y_res = 512
sample_rate = 22050
n_fft = 2048
hop_length = 512