Sloppy Butchery is a collection of Colab notebooks for various audio processing tools.
Sloppy Butcher is an audio power-chopper and randomizer. It takes a directory of audio files, chops it, shuffles it, and frankensteins it up into a single audio file according to your BPM, effects and other settings.
| Source description | Result |
|---|---|
| Random full length african choir-like songs. | MP3 |
| Random clips of Neurofunk tracks. | WAV |
| Random full length songs. | MP3 |
| Random full length finnish pop songs. | MP3 |
Autotune, surprisingly enough, autotunes audio files or directories of audio files to a) nearest note, b) a scale, c) another audio file or d) Chords Guru Turbo 100a Deluxe chord progression.
| Source description | Result |
|---|---|
| Original | WAV |
| Autotuned flat C | WAV |
| Autotuned to 1994 Nokia Tune | WAV |
| Autotuned to a chord progression | WAV |
Restores audio quality of a low sample rate audio a tiny bit by traditional methods.
audioldm-demo.mp4
Stuttering Butcher applies stutter edits and glitch on an audio file according to settings. Works by repeating N grains of odd zero-crossings of the audio signal — does not account for tempo, rhythmic patterns or "grid".
| Original audio | Stuttered | Glitched |
|---|---|---|
| WAV | WAV | WAV |
Poetic Butcher is a voice isolator and speech-to-words slicer. It isolates voice from audio source using Deezer Spleeter (convolutional neural network) and/or slices it to individual words according to settings using Mozilla Deepspeech (recurrent neural network).
Senior Butcher is a beat-slicer that improves the beat tracking precision of audio analysis libraries such as Librosa, Essentia or Aubio, particularly for the purpose of beat slicing.
| Track | Librosa | Butcher, ours |
|---|---|---|
| Aphex Twin - Windowlicker | MP3 | MP3 |
| Boards of Canada - Peacock Tail | MP3 | MP3 |
| 2Unlimited - No Limit | MP3 | MP3 |
-
Drum track extraction for timing: Stem track of drums is extracted from input audio by Deezer Spleeter (convolutional neural network) to be used as a timing track. Timing track is used for initial estimation of beat positions, i.e. input audio track (audiofile#1) is eventually sliced according to analysis performed on the timing track (audiofile#2).
-
Duration refinements by lossy compression of beat interval distribution: After initial beat tracking is performed according to timing track, beat durations are calculated from beat position distribution. Significant duration range is constricted by finding the modal value of approximated beat intervals. Currently the minimum beat duration within the significant range is eventually used as new absolute beat duration, which may increase the track tempo by a fraction of a bpm. This is likely to be optimized in the future.
-
Tail peak cluster shaving: Marginally prolonged beat duration outside the significant range is used for detection of high concentrations of peaks at the tail. These peak clusters are assumed subsequent beat beginnings. Should such peak cluster occur, beat start position is nudged backwards in time until the cluster at hand is positioned in the subsequent slice of a beat.
Audio Concat takes a directory of audio files and glues them together into one or more WAV files. This notebook was originally created to concatenate thousands of very short audio files produced by neural networks, and basically this notebook sparked the idea to create all the other notebooks in this repository.
Sloppy Noto is a data audiolizer residing in it's own repository. It turns numeric data into audio by interpreting it as digital audio signal sample magnitudes.
AudioLDM is a neural text-to-audio tool. Think Stable Diffusion but for audio.
Tortoise TTS is a neural text-to-speech (audio) with one notable difference: you also give it an audio input of a voice, and it will produce audio with the given voice.