soundfactory

soundfactory is a simple tool to experiment and be creative with audio taking a data-oriented approach. It is primarily directed to digital artists that want to generate audio samples from diverse data sources, and to audio-engineering students at their first steps into the beautiful world of additive sound synthesis.

The peculiarity about soundfactory is the adoption of fourier-series approximation when square, sawtooth, or triangle wave shapes are chosen. In fact, being approximated the single-oscillator signals will contain upper harmonics with respect to the analytical signal (with the same amplitude, frequency, and wave shape), that can be obtained by scipy (see tests). The result will be a warmer and unexpectedly coloured sound.

It comes from some scripts I have been writing a few years back and thought about having few reusable core modules and a command-line tool for researching new sounds. And I wanted to be able to develop it and have fun with it in my Jupyter Notebook of course!

You can find the documentation at https://soundfactory.readthedocs.io

---

WARNING

Before installing with pip check the pre-requirements

---

Examples

Random Sample Generator

from random import uniform, choice
from soundfactory import SignalBuilder, SignalPlotter
import matplotlib.pyplot as plt

n_components = 4
MAX_FREQ = 1000.

fname = ''
freqs, amps, shapes = list(), list(), list()
for _ in range(n_components):
    freq = round(uniform(20., MAX_FREQ), 1)
    freqs.append(freq)
    amp = round(uniform(0.1, 1.), 1)
    amps.append(amp)
    shape = choice(['sine', 'square', 'sawtooth', 'triangle'])
    shapes.append(shape)
    fname += '_'.join([str(freq), str(amp), shape])

s = SignalBuilder(
    freqs, amps, shapes,
    samplerate=96000,
    duration=2.
)
fname += '.wav'
s.export(fname)

# Analyse with SignalPlotter
sig = SignalPlotter(plt, fname, True)
sig.show(wmsec=0.1)

Reproducing a Sound Characteristic at an arbitrary Frequency

from numpy import where, sqrt
import matplotlib.pyplot as plt

from soundfactory import SignalBuilder, SignalPlotter
from soundfactory import Signal
from soundfactory.utils.scale import build_24_tet_scale

SCALE_INIT = {'E0': 20.6}

ref_label = list(SCALE_INIT.keys())[-1]
ref_frequency = list(SCALE_INIT.values())[-1]
scale_24 = build_24_tet_scale(ref_label, ref_frequency, max_octave=3)

ref_signal = Signal('/<path-to-packages>/soundfactory/samples/A3-Calib-220.wav')

freqs = ref_signal.SPECTRA['ch1_fft'][ref_signal.FREQUENCIES]
pws = ref_signal.SPECTRA['ch1_fft'][ref_signal.POWERS]

# Select only few of the spectral components
select_idx = where(pws > 0.00001)[0]
# Get the fundamental frequency to obtain the upper-harmonic orders
fundamental_idx = where(pws == pws.max())[0]
freq_ratios = freqs / freqs[fundamental_idx]

tone = 'E3𝄲'  # Choose an arbitrary tone NOT playble on the keyboard
input_freqs = scale_24[tone] * freq_ratios[select_idx]
input_amps = sqrt(2 * pws[select_idx])

my_signal = SignalBuilder(input_freqs,
                          input_amps, [
                          'sine' for _ in range(len(select_idx))]
                         )

filename = './%s_sample.wav' % tone
my_signal.export(filename)

# Analyse with SignalPlotter
sig = SignalPlotter(plt, filename, True)
sig.show(wmsec=0.1)