muse-python

Python implementation of Mouse Ultrasonic Source Estimator, a MATLAB tool developed to identify the location of a sound from microphone array data in the paper (Neunebel et al., 2015).

Installation

MUSE can be easily installed from PyPI as follows:

$ pip install muse-python

or from the latest code on GitHub with:

$ pip install git+https://github.com/ralphpeterson/muse-python

Overview

The main functions of this library are r_est_naive and r_est_jackknife. The function r_est_naive calculates the Reduced Steered Response Power (RSRP) at a grid of points in an arena, returning both the location of the maximum RSRP value as well as the grid itself. For more on RSRP values, see the original Neunebel et al. 2015 paper, which describes the calculation and rationale in detail.

r_est_jackknife builds on this by calculating multiple location estimates as in (Warren et al. 2018). For each vocal signal, r_est_jackknife systematically excludes one microphone, calculating a point estimate using only the data from the other microphones. This process is repeated for each mic, leading to a group of location estimates that are then averaged.

Getting started

For a detailed example, see examples/pyroomacoustics_muse_usage.ipynb, where we use the library pyroomacoustics to simulate the microphone data that would be recieved from a given audio source.

The following is a simple quickstart with randomly generated 'microphone input'.

import numpy as np
from muse import r_est_jackknife

# generate some fake audio data
n_mics = 4
N = 10000  # number of samples
fs = 100000  # sample rate, in Hz
rng = np.random.default_rng(seed=2022)
v = rng.random(size=(n_mics, N))

temp = 20  # temperature of the room in Celsius
f_lo = 0  # minimum frequency considered, in Hz
f_hi = fs/2  # max frequency considered, in Hz

# spatial parameters
x_len, y_len = (1, 2) # arbitrary dimensions of the room
res = 0.025  # desired resolution of grid at which RSRP values are calculated
# array representing locations of each mic in Cartesian coords
mic_pos = np.array([
    [0.2, 0.2, 1],  # units in meters
    [0.2, 0.8, 1],
    [1.8, 0.2, 1],
    [1.8, 0.8, 1]
])

avg_est, r_ests, rsrp_grids = r_est_jackknife(v, fs, f_lo, f_hi, temp, x_len, y_len, res, mic_pos)

print(f'--- Averaged Estimate --- \n{avg_est}')
print('--- Point Estimates --- ')
for i, est in enumerate(r_ests):
    print(f'Mic {i} removed:\n {est}')

Note that avg_est is the averaged result, r_ests is an array containing each point estimate, and rsrp_grids stores the full grid of RSRP values associated with each point estimate.

Out:

--- Averaged Estimate --- 
[[0.6474359 ]
 [1.12658228]]
--- Point Estimates --- 
Mic 0 removed:
 [[0.64102564]
 [1.56962025]]
Mic 1 removed:
 [[0.53846154]
 [1.36708861]]
Mic 2 removed:
 [[0.8974359 ]
 [0.48101266]]
Mic 3 removed:
 [[0.51282051]
 [1.08860759]]