Pyroomacoustics is a software package aimed at the rapid development and testing of audio array processing algorithms. The content of the package can be divided into three main components:
- Intuitive Python object-oriented interface to quickly construct different simulation scenarios involving multiple sound sources and microphones in 2D and 3D rooms;
- Fast C++ implementation of the image source model and ray tracing for general polyhedral rooms to efficiently generate room impulse responses and simulate the propagation between sources and receivers;
- Reference implementations of popular algorithms for STFT, beamforming, direction finding, adaptive filtering, source separation, and single channel denoising.
Together, these components form a package with the potential to speed up the time to market of new algorithms by significantly reducing the implementation overhead in the performance evaluation step. Please refer to this notebook for a demonstration of the different components of this package.
Room Acoustics Simulation¶
Consider the following scenario.
Suppose, for example, you wanted to produce a radio crime drama, and it so happens that, according to the scriptwriter, the story line absolutely must culminate in a satanic mass that quickly degenerates into a violent shootout, all taking place right around the altar of the highly reverberant acoustic environment of Oxford’s Christ Church cathedral. To ensure that it sounds authentic, you asked the Dean of Christ Church for permission to record the final scene inside the cathedral, but somehow he fails to be convinced of the artistic merit of your production, and declines to give you permission. But recorded in a conventional studio, the scene sounds flat. So what do you do?
—Schnupp, Nelken, and King, Auditory Neuroscience, 2010
Faced with this difficult situation, pyroomacoustics can save the day by simulating the environment of the Christ Church cathedral!
At the core of the package is a room impulse response (RIR) generator based on the image source model that can handle
- Convex and non-convex rooms
- 2D/3D rooms
The core image source model and ray tracing modules are written in C++ for better performance.
The philosophy of the package is to abstract all necessary elements of an experiment using an object-oriented programming approach. Each of these elements is represented using a class and an experiment can be designed by combining these elements just as one would do in a real experiment.
Let’s imagine we want to simulate a delay-and-sum beamformer that uses a linear array with four microphones in a shoe box shaped room that contains only one source of sound. First, we create a room object, to which we add a microphone array object, and a sound source object. Then, the room object has methods to compute the RIR between source and receiver. The beamformer object then extends the microphone array class and has different methods to compute the weights, for example delay-and-sum weights. See the example below to get an idea of what the code looks like.
The Room class also allows one to process sound samples emitted by sources, effectively simulating the propagation of sound between sources and microphones. At the input of the microphones composing the beamformer, an STFT (short time Fourier transform) engine allows to quickly process the signals through the beamformer and evaluate the output.
In addition to its core image source model simulation, pyroomacoustics also contains a number of reference implementations of popular audio processing algorithms for
- Short time Fourier transform (block + online)
- direction of arrival (DOA) finding
- adaptive filtering (NLMS, RLS)
- blind source separation (AuxIVA, Trinicon, ILRMA, SparseAuxIVA, FastMNMF)
- single channel denoising (Spectral Subtraction, Subspace, Iterative Wiener)
We use an object-oriented approach to abstract the details of specific algorithms, making them easy to compare. Each algorithm can be tuned through optional parameters. We have tried to pre-set values for the tuning parameters so that a run with the default values will in general produce reasonable results.
Install the package with pip:
pip install pyroomacoustics
A cookiecutter is available that generates a working simulation script for a few 2D/3D scenarios:
# if necessary install cookiecutter pip install cookiecutter # create the simulation script cookiecutter gh:fakufaku/cookiecutter-pyroomacoustics-sim # run the newly created script python <chosen_script_name>.py
We have also provided a minimal Dockerfile example in order to install and run the package within a Docker container. Note that you should increase the memory of your containers to 4 GB. Less may also be sufficient, but this is necessary for building the C++ code extension. You can build the container with:
docker build -t pyroom_container .
And enter the container with:
docker run -it pyroom_container:latest /bin/bash
The minimal dependencies are:
numpy scipy>=0.18.0 Cython pybind11
Cython is only needed to benefit from the compiled accelerated simulator.
The simulator itself has a pure Python counterpart, so that this requirement could
be ignored, but is much slower.
On top of that, some functionalities of the package depend on extra packages:
samplerate # for resampling signals matplotlib # to create graphs and plots sounddevice # to play sound samples mir_eval # to evaluate performance of source separation in examples
requirements.txt file lists all packages necessary to run all of the
scripts in the
This package is mainly developed under Python 3.5. We try as much as possible to keep things compatible with Python 2.7 and run tests and builds under both. However, the tests code coverage is far from 100% and it might happen that we break some things in Python 2.7 from time to time. We apologize in advance for that.
Under Linux and Mac OS, the compiled accelerators require a valid compiler to be installed, typically this is GCC. When no compiler is present, the package will still install but default to the pure Python implementation which is much slower. On Windows, we provide pre-compiled Python Wheels for Python 3.5 and 3.6.
Here is a quick example of how to create and visual the response of a beamformer in a room.
import numpy as np import matplotlib.pyplot as plt import pyroomacoustics as pra # Create a 4 by 6 metres shoe box room room = pra.ShoeBox([4,6]) # Add a source somewhere in the room room.add_source([2.5, 4.5]) # Create a linear array beamformer with 4 microphones # with angle 0 degrees and inter mic distance 10 cm R = pra.linear_2D_array([2, 1.5], 4, 0, 0.1) room.add_microphone_array(pra.Beamformer(R, room.fs)) # Now compute the delay and sum weights for the beamformer room.mic_array.rake_delay_and_sum_weights(room.sources[:1]) # plot the room and resulting beamformer room.plot(freq=[1000, 2000, 4000, 8000], img_order=0) plt.show()
A couple of detailed demos with illustrations are available.
A comprehensive set of examples covering most of the functionalities
of the package can be found in the
examples folder of the GitHub
How to contribute¶
If you would like to contribute, please clone the repository and send a pull request.
For more details, see our CONTRIBUTING page.
This package was developed to support academic publications. The package contains implementations for DOA algorithms and acoustic beamformers introduced in the following papers.
- H. Pan, R. Scheibler, I. Dokmanic, E. Bezzam and M. Vetterli. FRIDA: FRI-based DOA estimation for arbitrary array layout, ICASSP 2017, New Orleans, USA, 2017.
- I. Dokmanić, R. Scheibler and M. Vetterli. Raking the Cocktail Party, in IEEE Journal of Selected Topics in Signal Processing, vol. 9, num. 5, p. 825 - 836, 2015.
- R. Scheibler, I. Dokmanić and M. Vetterli. Raking Echoes in the Time Domain, ICASSP 2015, Brisbane, Australia, 2015.
If you use this package in your own research, please cite our paper describing it.
R. Scheibler, E. Bezzam, I. Dokmanić, Pyroomacoustics: A Python package for audio room simulations and array processing algorithms, Proc. IEEE ICASSP, Calgary, CA, 2018.
Copyright (c) 2014-2018 EPFL-LCAV Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
Table of contents¶
- `0.4.2`_ - 2020-09-24
- 0.4.1 - 2020-07-02
- `0.4.0`_ - 2020-06-03
- 0.3.1 - 2019-11-06
- 0.3.0 - 2019-11-06
- 0.2.0 - 2019-09-04
- 0.1.23 - 2019-04-17
- 0.1.22 - 2019-04-11
- 0.1.21 - 2018-12-20
- 0.1.20 - 2018-10-04
- 0.1.19 - 2018-09-24
- 0.1.18 - 2018-04-24
- 0.1.17 - 2018-03-23
- 0.1.16 - 2018-03-06
- 0.1.15 - 2018-02-23
- 0.1.14 - 2018-02-20
- Room Simulation
- Materials Database
- Dataset Wrappers
- Adaptive Filtering
- Blind Source Separation
- Direction of Arrival
- Single Channel Denoising
- Phase Processing
- Pyroomacoustics API