spectrs 0.2.0

Fast spectrogram creation library
Documentation

spectrs

Crates.io License

Fast spectrogram creation library.

Table of Contents

What It Does

spectrs is a Rust library for creating spectrograms from audio files. It provides:

  • STFT (Short-Time Fourier Transform) spectrograms with power or magnitude scaling
  • Mel-scaled spectrograms using HTK or Slaney scales
  • Audio I/O for reading WAV files (mono/stereo, 8/16/32-bit)
  • Resampling to arbitrary sample rates
  • Image export of spectrograms with multiple colormaps (Viridis, Magma, Inferno, Plasma, Gray)
  • Compatibility with librosa

Installation

As a Command-Line Tool

Install the binary with full features (CLI + image support):

cargo install spectrs

As a Library

When using spectrs as a library dependency, you typically want minimal dependencies. Add it with:

# Minimal dependencies (core audio and spectrogram functionality only)
cargo add spectrs --no-default-features

# With image export support
cargo add spectrs --no-default-features --features image

Quick Start

Command-Line Usage

After installing with cargo install spectrs, you can process audio files from the command line. The CLI will automatically create PNG images with the same name as the input file(s).

# Process a single file with default settings
spectrs audio.wav

# Create a mel spectrogram with 128 mel bands
spectrs audio.wav --n-mels 128

# Customize spectrogram parameters
spectrs audio.wav \
  --n-fft 2048 \
  --hop-length 512 \
  --n-mels 128 \
  --spec-type power \
  --colormap viridis

# Process all WAV files in a directory
spectrs audio_folder/

# See all available options
spectrs --help

Colormaps

spectrs supports multiple colormaps for spectrogram visualization, including viridis, magma, inferno, plasma and simply gray. All colormaps implementations are based on the matplotlib colormaps.

Librosa Compatibility

spectrs is designed to replicate librosa's behaviour in the Rust ecosystem. Compatibility tests are part of the test suite, ensuring correlation between librosa's and spectrs' above 99% and relative error on relevant bands < 2%.