Custom Noise Generator

A Python package for generating noise signals with custom power spectral density (PSD) of 1/f^alpha.

Installation

pip install custom_noise

Or install from source:

git clone https://github.com/commuted/custom_noise.git
cd custom_noise
pip install -e .

Quick Start

from custom_noise import pink_noise, brown_noise

# Generate 5 seconds of pink noise
signal = pink_noise(duration=5)

# Save brown noise to a WAV file
brown_noise(duration=10, output_file="brown_noise.wav")

Noise Types

Function	Alpha	Description
white_noise()	0	Equal power at all frequencies
pink_noise()	1	1/f spectrum, equal power per octave
brown_noise()	2	1/f² spectrum, deep rumbling (also called red/Brownian noise)*
blue_noise()	-1	f spectrum, bright/hissing quality
violet_noise()	-2	f² spectrum, high-frequency emphasis

*Note: True Brownian noise has infinite power at DC (0 Hz), causing unbounded drift. This implementation leaves the DC component unscaled to produce a bounded, stationary signal suitable for audio use.

Usage

Basic Usage

from custom_noise import pink_noise, white_noise, brown_noise

# Generate noise (returns numpy array)
signal = pink_noise(duration=3)

# Save directly to file
white_noise(duration=5, output_file="white.wav")

# Custom sample rate
signal = brown_noise(duration=2, sample_rate=48000)

Output Formats

from custom_noise import pink_noise

# 16-bit integers (default) - ready for WAV files
signal_int = pink_noise(duration=1, dtype="int16")

# Normalized float [-1.0, 1.0] - for DSP processing
signal_float = pink_noise(duration=1, dtype="float32")

Custom Alpha Values

For noise colors not covered by the convenience functions:

from custom_noise import generate_custom_noise

# Custom alpha value
signal = generate_custom_noise(alpha=0.5, duration=3)

# Gray noise (alpha ≈ 1, psychoacoustically balanced)
signal = generate_custom_noise(alpha=1, duration=5)

Reproducibility

from custom_noise import pink_noise

# Same seed = same output
signal1 = pink_noise(duration=1, random_seed=42)
signal2 = pink_noise(duration=1, random_seed=42)
assert (signal1 == signal2).all()

# Random each time
signal = pink_noise(duration=1, random_seed=None)

Visualization

from custom_noise import generate_custom_noise

# Plot the power spectral density
generate_custom_noise(alpha=1, duration=3, plot_psd=True)

Command Line Interface

# Generate pink noise
python -m custom_noise --type pink --duration 5 -o pink.wav

# Generate with custom alpha
python -m custom_noise --alpha 1.5 --duration 10 -o custom.wav

# Show PSD plot
python -m custom_noise --type brown --duration 3 --plot

# See all options
python -m custom_noise --help

CLI Options

Option	Description
-t, --type TYPE	Noise type: white, pink, brown, blue, violet
-a, --alpha ALPHA	Custom alpha exponent
-d, --duration SECONDS	Duration in seconds (default: 3)
-r, --sample-rate HZ	Sample rate in Hz (default: 44100)
-o, --output FILE	Output WAV file path
--plot	Display PSD plot
-s, --seed SEED	Random seed for reproducibility

API Reference

Convenience Functions

All convenience functions share these parameters:

Parameter	Type	Default	Description
duration	float	3	Duration in seconds
sample_rate	int	44100	Sample rate in Hz
output_file	str	None	Path to save WAV file
random_seed	int	42	Seed for reproducibility (None for random)
dtype	str	"int16"	Output type: "int16", "float32", "float64"

generate_custom_noise()

Full control over noise generation:

generate_custom_noise(
    sample_rate=44100,    # Sample rate in Hz
    duration=3,           # Duration in seconds
    alpha=None,           # PSD exponent (mutually exclusive with noise_type)
    noise_type=None,      # Preset: "white", "pink", "brown", "blue", "violet"
    output_file=None,     # Path to save WAV file
    plot_psd=False,       # Show PSD plot
    random_seed=42,       # Seed for reproducibility
    dtype="int16",        # Output format
)

How It Works

1. Generate Gaussian white noise
2. Apply FFT to transform to frequency domain
3. Scale frequencies by 1/|f|^(α/2) to achieve desired PSD slope
4. Inverse FFT back to time domain
5. Normalize and convert to requested output format

Requirements

• Python >= 3.9
• NumPy
• SciPy
• Matplotlib (for plotting)

License

MIT License