Randomized Image Method (RIM)

This project provides a Matlab implementation of the image method, a widely used acoustic model to calculate the room impulse response (RIR) of a rectangular room. The software allows multiple microphone positions and fractional delays.

If needed, the software generates can randomise the position of the image sources to avoid the occurrence of sweeping echoes, resulting in the so-called randomized image method. See De Sena et al. "On the modeling of rectangular geometries in room acoustic simulations." IEEE/ACMT TASLP (2015).

For the Julia implementation see https://github.com/nantonel/RIM.jl, maintained by Niccolò Antonello.

Getting Started

Prerequisites and installation

No installation is needed. No toolbox needed. This version has been tested on Matlab2020a, but I have no reason to believe it wouldn't work on previous versions too.

Running the tests

To run the tests simply run rim_tests

Using the software

The main script is rim.m:

RIM(mic_pos, source_pos, room_dim, beta, rir_length, Fs)

• mic_pos is the 3xM matrix with the position of M omnidirectional microphones. The dimensions are in [meters].
• source_pos is the 3x1 vector with the position of the omni sound source [meters].
• room_dim is the 3x1 vector with the dimensions of the room [meters].
• beta is the 2x3 vector with the reflection coefficient of the walls, in this order: [x1, y1, z1; x2, y2, z2], with e.g. x1 and x2 are the reflection coefficient of the surface orthogonal to the x-axis and with the subscript 1 referring to walls adjacent to the coordinate origin, and the subscript 2 referring to the opposite wall. In the anechoic case beta=zeros(2,3).
• rir_length is the length of the RIR [seconds].
• Fs is the sampling frequency [Hz].

The following are optional inputs:

RIM(mic_pos, source_pos, room_dim, beta, rir_length, Fs, rand_dist, Tw, Fc, c)

• rand_dist is the random distance added to the position of the image sources in [meters]; in the Transaction paper we found that rand_dist=0.08 m was sufficient to remove sweeping echoes in most cases studied (rand_dist=0 for the standard image method) (default is 0 cm).
• Tw is the length of the fractional delay filter in [seconds] (default is Tw=0.004 s)
• Fc is the cut-off frequency of the fractional delay filter in [Hz] (default is Fc=0.9*(Fs/2))
• c is the speed of sound in m/s (default is c=343)

Multiple microphone positions

As opposed to the original implementation in the paper, this implementation of the RIM allows to return the room impulse response at multiple positions. This is done by using the same randomised position of the image source for all receivers. The wavefront due to each image source would be originating in a slightly inaccurate point (which is necessary to remove sweeping echoes), but the relative time arrivals and directions would be correct across receivers. This allows application in studies simulating multiple omnidirectional microphones, e.g. beamforming.

Generating the plots in the 2015 Transaction paper

Run rim_taslp_plots.

Note with regard to Figure 11a (Highly regular setup with Randomized Image Method): this repository uses a different randomisation strategy compared to the paper. While in the paper the source was moved on a line between the original image sound and the observer by +-8 cm (i.e. 16 cm spread), here, the source is moved within a cube with edge 16 cm and centered at the original image source. Visual inspection suggests the effect is similar. If you'd like to replicate this figure exactly as in the Transaction paper please refer to desena.org/sweep/.

Contributing

Contributions are welcome.

Author

• Enzo De Sena - desena.org

License

This project is licensed under the MIT License - see the LICENSE file for details

If you use this code, please cite the following paper:

De Sena, E., Antonello, N., Moonen, M. and Van Waterschoot, T., 2015. On the modeling of rectangular geometries in room acoustic simulations. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 23(4), pp.774-786.