CRUSADE

Conversion of Raw-audio Using Spikes Analog Digital Encoders

This repository contains implementations of different analogue-to-spike converters for raw audio signal.

Features

This project contains Python implementations of various methods and neural models used to transform audio signals into spike/event trains, using JAX. It includes: - Standalone ADM. - Filterbank with ADM. - Filterbank with Resonate and Fire neurons. - Filterbank with phase encoding (under development).

Installation

Recommended version for installation is using uv

uv sync --frozen --all-extras

Or with pip

pip install ".[dev]"

Example

Example for use filterbank_ADM:

import jax.numpy as jnp
import matplotlib.pyplot as plt
from scipy.signal import chirp

from crusafe.conversion_methods import filterbank_ADM

sr = 44100
duration = 0.1
t = jnp.arange(int(sr * duration)) / sr
audio = chirp(t, f0=200, f1=2000, t1=duration, method="linear")

fb = filterbank_ADM(
    num_neurons=16, freq_min=200, freq_max=2000, freq_distribution="linear"
)
event_time, event_address, event_magnitude = fb(audio, sampling_rate=sr)

plt.figure()
plt.scatter(x=event_time, y=event_address, c=event_magnitude, cmap="bwr")
plt.show()

Tests

• On pc

To check if the models are working:

uv run pytest

or if you are in the envirnoment

pytest

To check before commit:

uv run pre-commit run --all-files

to install the pre-commit (only forst time):

uv run pre-commit install

once installed it runs automatically for every commit

• On git

It does automatically runs all the tests