title	author
polar-measurement	Tony Terrasa, David Ramsay

polar-measurement

This module contains scripts for taking polar measurements as well as well as submodules for creating and manipulating microphone arrays and optimizing filters to achieve a desired response in tensorflow

Installation

This package requires python

It is recommended that you use a virtual environment so that it does not interfere with your local version of Python or numpy. From within your virual environment and in the polar-measurement folder:

$ pip3 install -r requirements.txt

Next, I recommend that you link the submodules of polar-measurement to the site-packages directory in your virtual environment or to your PYTHONPATH

$ ln -s /path_to_polar-measurement/pythonAudioMeasurements /path_to_virtualenv/lib/python3.6/site-packages/
$ ln -s /path_to_polar-measurement/pyAudioFilter /path_to_virtualenv/lib/python3.6/site-packages/

You should now be able to use the packages. You can test by running the following without errors

$ python3
Python 3.6.9 (default, Apr 18 2020, 01:56:04) 
[GCC 8.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from pythonAudioMeasurements.audioSample import audioSample
>>> from pyAudioFilter.polar_optimizer import PolarOptimizer

Usage

There are examples show in the submodule of how to use individual submodules. Below is information on how to use them together, namely to optimize filters for a microphone array to achieve a stop-band in theta and frequency

First, the necessary imports:

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
from pythonAudioMeasurements.polarData import polarData
from pythonAudioMeasurements.microphone import Microphone
from pythonAudioMeasurements.microphoneArray import MicrophoneArray
from pyAudioFilter.polar_optimizer import PolarOptimizer

Next, we should create a MicrophoneArray of the same microphone from the data/19_Jan15_fixedpkls folder.

Load in the a microphone as a polarData object.

filename = "/path_to_polar-measurement/data/19_Jan15_fixedpkls/spv1840.pkl" 
pd = polarData.fromPkl(filename)

Now, set locations, create Microphone instances and store them in a MicrophoneArray instance.

# set specific locations for the mics, given as x,y in mm
locations = [(0,0), (100,200), (-100, 200), (-100,-200), (100,-200)]

# create a microphone array
mic_array =  MicrophoneArray([Microphone(pd, loc) for loc in locations])

Alternatively, it you would like to randomly generate locations of many microphones:

num_mics = 20
box_width = 600 # mm
locations = box_width*np.random.rand(num_mics,2)

These instances can now be used to visualize the Microphone or MicrophoneArray.

To do: add a way to quickly show the polarData response at certain frequencies

mic_array.visualize() # plots the microphone locations

However, these representations need to formated to be used in a tensorflow model.

# return np.array's
angles, freqs, mic_responses = mic_array.tf_prep()

angles = tf.constant(angles)
freqs = tf.constant(freqs)
mic_responses = [tf.constant(mic) for mic in mic_responses]
fs = tf.constant(44.1e3, dtype=freqs.dtype)

po = PolarOptimizer(angles, freqs, mic_responses, fs=fs, learning_rate=0.01)

Now we set a target angle and frequencies for our stop-band

stop_band_theta = [45, 135]
target_range_freq = [500, 1000]

And, we are ready to run our optimization

losses = []
NUM_EPOCHS = 60

for epoch in range(NUM_EPOCHS):

    # run a step
    po.train(stop_band_theta, target_range_freq, threshold=-20.)
    losses.append(po.current_loss.numpy()) # keep track of the loss over the epochs

    if epoch % 10 == 0:
        pd = po.to_polar_data()
        pd.plotFreqs(np.linspace(target_range_freq[0], target_range_freq[1], 5),\
            title="epoch: %d, loss: %.2f"%(epoch, losses[-1]))

plt.figure(2)
plt.plot(losses)
plt.title("Loss over Epoch. Stop band: theta=(%d,%d), f=(%.2f, %.2f)"%(\
    10,60,500,1000))

pd_final = po.to_polar_data()
pd_final.plotFreqs(np.linspace(target_range_freq[0], target_range_freq[1], 5), show=False)

plt.show()

In this case pd_final is now a polarData representation of the resulting microphone array.

A full version of this script can be found in the sample_optimization.py

Future Work

use collected baseline anechoic chamber measurements to better mic response fidelity

dramsay9/polar-measurement

polar-measurement

Installation

Usage

Future Work