spectrs

Fast spectrogram creation library.

What It Does
Installation
Quick Start
Features
Examples
Librosa Compatibility
Performance
API Reference

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
Compatibility with librosa

Installation

Add spectrs to your project with

cargo add spectrs

or with image export support

cargo add spectrs --features image

Examples

Basic STFT Spectrogram

use spectrs::io::audio::read_audio_file_mono;
use spectrs::spectrogram::stft::{par_compute_spectrogram, SpectrogramType};

let (audio, _sr) = read_audio_file_mono("speech.wav")?;

// Compute magnitude spectrogram
let spec = par_compute_spectrogram(
    &audio,
    2048,   // n_fft: higher = better frequency resolution
    512,    // hop_length: lower = better time resolution
    2048,   // win_length: window size
    true,   // center: pad the signal
    SpectrogramType::Magnitude
);

Mel Spectrogram

use spectrs::spectrogram::mel::{convert_to_mel, MelScale};

// First compute STFT, then convert to mel
let mel_spec = convert_to_mel(
    &spec,
    22050,              // sample_rate
    2048,               // n_fft
    128,                // n_mels: number of mel bands
    Some(20.0),         // f_min: lowest frequency
    Some(8000.0),       // f_max: highest frequency
    MelScale::Slaney    // librosa default
);

Save Spectrogram as Image

use spectrs::io::image::save_spectrogram_image;

// Requires "image" feature
save_spectrogram_image(&mel_spec, "spectrogram.png")?;

Single-Threaded Mode

For small files or embedded systems:

use spectrs::spectrogram::stft::compute_spectrogram;

// Non-parallelized version (lower overhead)
let spec = compute_spectrogram(
    &audio,
    512,
    160,
    400,
    false,
    SpectrogramType::Power
);

Complete Pipeline

use spectrs::io::audio::{read_audio_file_mono, resample};
use spectrs::spectrogram::stft::{par_compute_spectrogram, SpectrogramType};
use spectrs::spectrogram::mel::{convert_to_mel, MelScale};

// Read and process audio file
let (audio, sr) = read_audio_file_mono("song.wav")?;

// Resample to standard rate
let audio = resample(audio, sr, 22050)?;

// Compute spectrogram
let spec = par_compute_spectrogram(&audio, 2048, 512, 2048, true, SpectrogramType::Power);

// Convert to mel
let mel = convert_to_mel(&spec, 22050, 2048, 128, None, None, MelScale::HTK);

// Save as image (requires "image" feature)
#[cfg(feature = "image")]
spectrs::io::image::save_spectrogram_image(&mel, "output.png")?;

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%.

Run compatibility tests:

cargo test --test test_librosa_compatibility

spectrs 0.1.1