spectrofast 0.2.0

Fast FFTW-based spectrogram computation — a drop-in replacement for scipy.signal.spectrogram

spectrofast

Fast FFTW-based spectrogram computation for Python. A simple, fast alternative to scipy.signal.spectrogram.

Installation

Prerequisites

spectrofast requires the FFTW3 library and a C compiler.

Windows:

• Install Visual Studio Build Tools (select "Desktop development with C++")
• FFTW is bundled with the package, no extra steps needed

Linux (Debian/Ubuntu):

sudo apt install libfftw3-dev

Linux (Fedora):

sudo dnf install fftw-devel

macOS:

brew install fftw

Install spectrofast

pip install spectrofast

Or install from source:

git clone https://github.com/yourusername/spectrofast.git
cd spectrofast
pip install .

Usage

import numpy as np
from spectrofast import spectrogram

# Single signal
signal = np.sin(2 * np.pi * np.arange(1024) / 32).astype(np.float64)
result = spectrogram(signal, nperseg=256, noverlap=128)
print(result.shape)  # (4, 129)

# Batch of signals
signals = np.random.randn(100, 1024).astype(np.float64)
result = spectrogram(signals, nperseg=256, noverlap=128)
print(result.shape)  # (100, 4, 129)

# With zero-padding
result = spectrogram(signal, nperseg=256, noverlap=128, nfft=512)
print(result.shape)  # (4, 257)

Output

The spectrogram function returns magnitude-squared values (i.e., real^2 + imag^2). This avoids the expensive sqrt operation and is suitable for most spectral analysis tasks.

License

MIT