csteinmetz1 / pymixconsole

Headless multitrack mixing console in Python
https://csteinmetz1.github.io/pymixconsole
MIT License
116 stars 13 forks source link

pymixconsole pymixconsole

Headless multitrack mixing console in Python

Installation

pip install git+https://github.com/csteinmetz1/pymixconsole

Usage

Setup a mixing console with a set of tracks from a multitrack project and apply processing per block. By default, a console will contain n channels and each channel will have a series of default processors:

gain -> polarity inverter -> parametric EQ -> compressor -> gain (fader) -> stereo panner

These are setup in such a way that if you do not modify their settings the signal should pass largely unprocessed. Additionally a console is initialized with two effect busses, one for reverb and one for delay. Finally there is a master bus which sums the output of all the busses and channels and then applies a simple processing chain:

parametric EQ -> compressor

In the example below you can see how to initialize a console and then pass multitrack data into the console and process it block by block to get the output.

Basic processing

One way to apply processing is to create a multidimensional array of shape [samples, tracks/channels], where each channel is a mono stream of audio, which will be processed by the associated channel in the console.

In this example we create an array with 8 channels of audio and then instantiate a default console with 8 channels. Then we iterate over the input data by the block_size and we pass each block to the console's process_block() function, which takes this array, applies each channel processor, and return a stereo mix. We then store this output in our pre-allocated array. We finally save this data to a .wav file with pySoundFile as the end.

import numpy as np
import soundfile as sf
import pymixconsole as pymc

data = np.random.rand(44100,8)   # one second of audio for 8 mono tracks
rate = 44100                     # 44.1 kHz sampling rate
block_size = 512                 # processor block size

# create a mix console with settings that match our audio data
console = pymc.Console(block_size=block_size, sample_rate=rate, num_channels=8)

# array to hold the output of the console (stereo)
out = np.empty(shape=(data.shape[0], 2))

# iterate over each block of data
for i in range(data.shape[0]//block_size):

    start = i * block_size 
    stop  = start + block_size

    out[start:stop,:] = console.process_block(data[start:stop,:])

# save out the processed audio
sf.write("output.wav", out, rate)

Console control

pymixconsole provides a high level of control over how the mix console is set up. By default, a console will include the supplied number of channels, as well as two busses (one for reverb, one for delay) and a master bus which features a compressor and equalizer. By default each channel is created with a pre-gain, polarity inverter, equaliser, compressor, post-gain, and a panner.

There are three levels of processors for each channel: pre-processors, core-processors, and post-processors. The distinction is useful since we want to impose some constraints on how these processors may be randomized in our randomize() method. The simple explanation is that the order of pre and post processors is never shuffled, while core-processors can be.

The defaults were chosen to be a good starting place for basic processing, but the user can customize this completely. For example, we can at any time add an extra processor to a channel as follows. Here we add a second compressor to the third channel's core-processors (zero-indexed), and then change the threshold parameter.

console.channels[2].processors.add(pymc.processors.Compressor(name="second-comp"))
console.channels[2].processor.get("second-comp").parameters.threshold.value = -22.0

Processor API

A number of basic processor units are included which can be included on a channel, bus, or the master bus.

Gain

Parameter Min. Max. Default Units Type Options
gain -80.0 24.0 0.0 dB float

Panner

Parameter Min. Max. Default Units Type Values
pan 0.0 1.0 0.5 float
outputs 2 2 2 outputs int
pan_law "-4.5dB" string "linear", "constant_power", "-4.5dB"

Equalizer

Parameter Min. Max. Default Units Type Values
low_shelf_gain -24.0 24.0 0.0 dB float
low_shelf_freq 20.0 1000.0 80.0 Hz float
first_band_gain -24.0 24.0 0.0 dB float
first_band_freq 200.0 5000.0 400.0 Hz float
first_band_q 0.1 10.0 0.7 float
second_band_gain -24.0 24.0 0.0 dB float
second_band_freq 500.0 6000.0 1000.0 Hz float
second_band_q 0.1 10.0 0.7 float
third_band_gain -24.0 24.0 0.0 dB float
third_band_freq 2000.0 10000.0 5000.0 Hz float
third_band_q 0.1 10.0 0.7 float
high_shelf_gain -24.0 24.0 0.0 dB float
high_shelf_freq 8000.0 20000.0 10000.0 Hz float

Delay

Parameter Min. Max. Default Units Type Values
delay 0 65536 5000 samples int
feedback 0.0 1.0 0.3 float
dry_mix 0.0 1.0 0.9 float
wet_mix 0.0 1.0 0.0 float

Compressor

Parameter Min. Max. Default Units Type Values
threshold -80.0 0.0 0.0 dB float
attack_time 0.001 500.0 10.0 ms float
release_time 0.0 1.0 100.0 ms float
ratio 1.0 100.0 2.0 float
makeup_gain -12.0 24.0 0.0 dB float

Algorithmic reverb

Parameter Min. Max. Default Units Type Values
room_size 0.1 1.0 0.5 float
damping 0.0 1.0 1.0 float
dry_mix 0.0 1.0 0.9 float
wet_mix 0.0 1.0 0.1 float
stereo_spread 0 100 23 int

Convolutional reverb

Parameter Min. Max. Default Units Type Values
dry_mix 0.0 1.0 0.9 float
wet_mix 0.0 1.0 0.1 float
decay 0.0 1.0 1.0 float
type "-4.5dB" string "sm-room", "md-room", "lg-room", "hall", "plate"

Cite

If you use this in your work please consider citing:

  @article{steinmetz2020mixing,
            title={Automatic multitrack mixing with a differentiable mixing console of neural audio effects},
            author={Steinmetz, Christian J. and Pons, Jordi and Pascual, Santiago and Serrà, Joan},
            journal={arXiv:2010.10291},
            year={2020}}