spafe-rs 0.1.1

Rust port of the spafe audio feature extraction library and jenellefeather/chcochleagram

Spafe

Native Rust implementation of the core spafe audio feature extraction APIs and jenellefeather/chcochleagram cochleagram routines.

The crate provides filter banks, spectrograms, cepstral features, spectral descriptors, frequency estimators, cochleagram generation, preprocessing helpers, and deterministic SVG visualization helpers.

Build

cargo build

The crate also has an opt-in portable SIMD path for hot numeric kernels. Because std::simd is still unstable, this feature currently requires nightly Rust:

cargo +nightly build --features portable-simd

Python Bindings

The crate exposes a PyO3 extension module through maturin. The Python package is published as spafe-rs and imports as spafe. Python functions use plain Python lists for one-dimensional signals and nested lists for matrices.

Install locally in a virtual environment:

python -m pip install maturin
maturin develop

Build a wheel:

python -m pip install build
python -m build --wheel

Basic Python usage:

import math
import spafe

fs = 16_000
sig = [math.sin(2.0 * math.pi * 440.0 * n / fs) for n in range(fs)]
opts = spafe.FeatureOptions(fs=fs, nfft=256, nfilts=24)

ceps = spafe.mfcc(sig, opts)
fbanks, centers = spafe.mel_filter_banks(spafe.FilterBankOptions(nfft=256))

print(len(ceps), len(ceps[0]))
print(len(fbanks), len(centers))

Test

cargo test --all-targets
cargo clippy --all-targets -- -D warnings
RUSTDOCFLAGS="-D warnings -D missing_docs" cargo doc --no-deps

Examples

Generate a filter-bank SVG:

cargo run --example fbanks
open target/fbanks.svg

Generate a cochleagram SVG:

cargo run --example cochleagram
open target/cochleagram.svg

Python examples live in examples/python, use matplotlib, and write outputs to target/python-examples:

python -m pip install matplotlib
python examples/python/features.py
python examples/python/fbanks.py
python examples/python/pitch.py
python examples/python/cochleagram.py

Interactive notebooks live in notebooks:

python -m pip install notebook matplotlib numpy
jupyter notebook notebooks

They cover feature extraction, filter banks, pitch tracking, cochleagrams, spectrogram variants, conversion helpers, parameter sensitivity, option tuning, performance, and visualization/export/classification recipes. The real-audio notebook uses notebooks/sample_uk.opus and requires ffmpeg.

Basic Usage

use spafe::prelude::*;

fn main() -> spafe::Result<()> {
    let fs = 16_000;
    let signal = vec![0.0; fs];
    let opts = FeatureOptions {
        fs,
        ..Default::default()
    };

    let coeffs = mfcc(&signal, &opts)?;
    println!("frames={}, coeffs={}", coeffs.nrows(), coeffs.ncols());

    let coch = cochleagram(&signal, &CochleagramOptions {
        signal_size: signal.len(),
        sr: fs,
        env_sr: 400,
        ..Default::default()
    })?;
    println!("filters={}, samples={}", coch.cochleagram.nrows(), coch.cochleagram.ncols());
    Ok(())
}