s/np/jnp/g s/py/jax/g

Docs below are from the original repo for the most part

jaxloudnorm

Flexible audio loudness meter in Python.

Implementation of ITU-R BS.1770-4.
Allows control over gating block size and frequency weighting filters for additional control.

For full details on the implementation see our paper with a summary in our AES presentation video.

Installation

Install from GitHub repo using

pip install git+https://github.com/boris-kuz/jaxloudnorm

Usage

Find the loudness of an audio file

It's easy to measure the loudness of a wav file. Here we use PySoundFile to read a .wav file as an ndarray.

import soundfile as sf
import jaxloudnorm as jln

data, rate = sf.read("test.wav") # load audio (with shape (samples, channels))
meter = jln.Meter(rate) # create BS.1770 meter
loudness = meter.integrated_loudness(data) # measure loudness

Loudness normalize and peak normalize audio files

Methods are included to normalize audio files to desired peak values or desired loudness.

import soundfile as sf
import jaxloudnorm as jln

data, rate = sf.read("test.wav") # load audio

# peak normalize audio to -1 dB
peak_normalized_audio = jln.normalize.peak(data, -1.0)

# measure the loudness first 
meter = jln.Meter(rate) # create BS.1770 meter
loudness = meter.integrated_loudness(data)

# loudness normalize audio to -12 dB LUFS
loudness_normalized_audio = jln.normalize.loudness(data, loudness, -12.0)

Advanced operation

A number of alternate weighting filters are available, as well as the ability to adjust the analysis block size. Examples are shown below.

import soundfile as sf
import jaxloudnorm as jln
from jaxloudnorm import IIRfilter

data, rate = sf.read("test.wav") # load audio

# block size
meter1 = jln.Meter(rate)                               # 400ms block size
meter2 = jln.Meter(rate, block_size=0.200)             # 200ms block size

# filter classes
meter3 = jln.Meter(rate)                               # BS.1770 meter
meter4 = jln.Meter(rate, filter_class="DeMan")         # fully compliant filters  
meter5 = jln.Meter(rate, filter_class="Fenton/Lee 1")  # low complexity improvement by Fenton and Lee
meter6 = jln.Meter(rate, filter_class="Fenton/Lee 2")  # higher complexity improvement by Fenton and Lee
meter7 = jln.Meter(rate, filter_class="Dash et al.")   # early modification option

# create your own IIR filters
my_high_pass  = IIRfilter(0.0, 0.5, 20.0, rate, 'high_pass')
my_high_shelf = IIRfilter(2.0, 0.7, 1525.0, rate, 'high_shelf')

# create a meter initialized without filters
meter8 = jln.Meter(rate, filter_class="custom")

# load your filters into the meter
meter8._filters = [my_high_pass, my_high_shelf]

Batched operation

Using jax allows us to calculate loudness and normalize across a batch dimension using vmap (or pmap across devices). Examples from tests:

def test_batched_integrated_loudness():
    data, rate = sf.read("tests/data/sine_1000.wav")
    meter = pyln.Meter(rate)
    loudness = jax.vmap(meter.integrated_loudness)(jnp.stack([data, data, data]))

    assert jnp.allclose(loudness, jnp.full(loudness.shape, -3.0523438444331137))

def test_batched_loudness_normalize():
    data, rate = sf.read("tests/data/sine_1000.wav")
    data = jnp.stack([data, data, data, data])
    meter = pyln.Meter(rate)
    loudness = jax.vmap(meter.integrated_loudness)(data)
    norm = jax.vmap(pyln.normalize.loudness, in_axes=(0, 0, None))(data, loudness, -6.0)
    loudness = jax.vmap(meter.integrated_loudness)(norm)

    assert jnp.allclose(loudness, jnp.full(loudness.shape, -6.0))

`

Dependencies

• SciPy (https://www.scipy.org/)
• NumPy (http://www.numpy.org/)
• JAX (https://jax.readthedocs.io/en/latest/index.html)
• jaxtyping (https://docs.kidger.site/jaxtyping/)
• python-soundfile (https://github.com/bastibe/python-soundfile)

References

Ian Dash, Luis Miranda, and Densil Cabrera, "Multichannel Loudness Listening Test," 124th International Convention of the Audio Engineering Society, May 2008

Pedro D. Pestana and Álvaro Barbosa, "Accuracy of ITU-R BS.1770 Algorithm in Evaluating Multitrack Material," 133rd International Convention of the Audio Engineering Society, October 2012

Pedro D. Pestana, Josh D. Reiss, and Álvaro Barbosa, "Loudness Measurement of Multitrack Audio Content Using Modifications of ITU-R BS.1770," 134th International Convention of the Audio Engineering Society, May 2013

Steven Fenton and Hyunkook Lee, "Alternative Weighting Filters for Multi-Track Program Loudness Measurement," 143rd International Convention of the Audio Engineering Society, October 2017

Brecht De Man, "Evaluation of Implementations of the EBU R128 Loudness Measurement," 145th International Convention of the Audio Engineering Society, October 2018.