Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/johnvinyard/zounds
Zounds is a dataflow library for building directed acyclic graphs that transform audio. It uses the featureflow library to define the processing pipelines.
https://github.com/johnvinyard/zounds
audio dsp machine-learning numpy processing-pipelines signal-processing sound
Last synced: about 1 month ago
JSON representation
Zounds is a dataflow library for building directed acyclic graphs that transform audio. It uses the featureflow library to define the processing pipelines.
- Host: GitHub
- URL: https://github.com/johnvinyard/zounds
- Owner: JohnVinyard
- License: mit
- Created: 2016-03-23T01:55:04.000Z (almost 9 years ago)
- Default Branch: master
- Last Pushed: 2022-12-08T00:34:47.000Z (about 2 years ago)
- Last Synced: 2024-10-15T17:43:45.580Z (2 months ago)
- Topics: audio, dsp, machine-learning, numpy, processing-pipelines, signal-processing, sound
- Language: Python
- Homepage:
- Size: 2.33 MB
- Stars: 23
- Watchers: 5
- Forks: 6
- Open Issues: 27
-
Metadata Files:
- Readme: README.md
- License: LICENSE.txt
Awesome Lists containing this project
README
[](https://travis-ci.org/JohnVinyard/zounds)
[](https://coveralls.io/github/JohnVinyard/zounds?branch=master)
[](https://pypi.python.org/pypi/zounds)
[](http://zounds.readthedocs.io/en/latest/?badge=latest)
[](https://opensource.org/licenses/MIT)
# Motivation
Zounds is a python library for working with sound. Its primary goals are to:
- layer semantically meaningful audio manipulations on top of numpy arrays
- [help to organize the definition and persistence of audio processing
pipelines and machine learning experiments with sound](https://github.com/JohnVinyard/zounds/tree/master/zounds/learn)
Audio processing graphs and machine learning pipelines are defined using
[featureflow](https://github.com/JohnVinyard/featureflow).
# A Quick Example
```python
import zounds
Resampled = zounds.resampled(resample_to=zounds.SR11025())
@zounds.simple_in_memory_settings
class Sound(Resampled):
"""
A simple pipeline that computes a perceptually weighted modified discrete
cosine transform, and "persists" feature data in an in-memory store.
"""
windowed = zounds.ArrayWithUnitsFeature(
zounds.SlidingWindow,
needs=Resampled.resampled,
wscheme=zounds.HalfLapped(),
wfunc=zounds.OggVorbisWindowingFunc(),
store=True)
mdct = zounds.ArrayWithUnitsFeature(
zounds.MDCT,
needs=windowed)
weighted = zounds.ArrayWithUnitsFeature(
lambda x: x * zounds.AWeighting(),
needs=mdct)
if __name__ == '__main__':
# produce some audio to test our pipeline, and encode it as FLAC
synth = zounds.SineSynthesizer(zounds.SR44100())
samples = synth.synthesize(zounds.Seconds(5), [220., 440., 880.])
encoded = samples.encode(fmt='FLAC')
# process the audio, and fetch features from our in-memory store
_id = Sound.process(meta=encoded)
sound = Sound(_id)
# grab all the frequency information, for a subset of the duration
start = zounds.Milliseconds(500)
end = start + zounds.Seconds(2)
snippet = sound.weighted[start: end, :]
# grab a subset of frequency information for the duration of the sound
freq_band = slice(zounds.Hertz(400), zounds.Hertz(500))
a440 = sound.mdct[:, freq_band]
# produce a new set of coefficients where only the 440hz sine wave is
# present
filtered = sound.mdct.zeros_like()
filtered[:, freq_band] = a440
# apply a geometric scale, which more closely matches human pitch
# perception, and apply it to the linear frequency axis
scale = zounds.GeometricScale(50, 4000, 0.05, 100)
log_coeffs = scale.apply(sound.mdct, zounds.HanningWindowingFunc())
# reconstruct audio from the MDCT coefficients
mdct_synth = zounds.MDCTSynthesizer()
reconstructed = mdct_synth.synthesize(sound.mdct)
filtered_reconstruction = mdct_synth.synthesize(filtered)
# start an in-browser REPL that will allow you to listen to and visualize
# the variables defined above (and any new ones you create in the session)
app = zounds.ZoundsApp(
model=Sound,
audio_feature=Sound.ogg,
visualization_feature=Sound.weighted,
globals=globals(),
locals=locals())
app.start(9999)
```
Find more inspiration in the [examples folder](https://github.com/JohnVinyard/zounds/tree/master/examples),
or on the [blog](http://johnvinyard.github.io/).
# Installation
## Libsndfile Issues
Installation currently requires you to build lbiflac and libsndfile from source, because of
[an outstanding issue](https://github.com/bastibe/PySoundFile/issues/130) that will be corrected when the apt package
is updated to `libsndfile 1.0.26`. Download and run
[this script](https://raw.githubusercontent.com/JohnVinyard/zounds/master/setup.sh) to handle this step.
## Numpy and Scipy
The [Anaconda](https://www.continuum.io/downloads) python distribution is highly recommended.
## Zounds
Finally, just:
```bash
pip install zounds
```