Complex rooms made with polygons and from corners freezes at room.image_source_model() for microphones with directivity

[whojavumusic]

I made an L shaped room and placed a Genelec_8020 source with HATS microphone given in the example. The code freezes when room.image_source_model() is called. Works fine with shoebox room.

[whojavumusic]

The code freezes in room.image_source_model() at
n_visible_sources = self.room_engine.image_source_model(source.position)

[fakufaku]

Thanks for catching this. Is it possible to share a code snippets that reproduce the problem?

[whojavumusic]

I realized that lower image orders (like 2) usually don't freeze, could it be a computation limitation? Anyway here's the snippet:

import numpy as np
import pyroomacoustics as pra
from pyroomacoustics.directivities import *

corners = np.array([[0,0], [0,5], [1,5], [1,2], [2,2], [2,5], [3,5], [3,0], [2,0], [2,1], [1,1], [1,0]]).T

room = pra.Room.from_corners(
corners,
fs=16000,
max_order=10,
)

room.extrude(3)

rot = Rotation3D([0, 0], "yz", degrees=True)
measurements = MeasuredDirectivityFile("LSPs_HATS_GuitarCabinets_Akustikmessplatz", fs=16000)

source_directivity = measurements.get_source_directivity('Genelec_8020',rot)
room.add_source([1.5,1.5,1.5], source_directivity)

hrtf = MeasuredDirectivityFile(path="mit_kemar_normal_pinna.sofa", fs=16000)
dir_left = hrtf.get_mic_directivity("left", orientation=rot)
dir_right = hrtf.get_mic_directivity("right", orientation=rot)
room.add_microphone([0.5,1.5,1.5], directivity=dir_left)
room.add_microphone([0.5,1.5,1.5], directivity=dir_right)

#freezes afterwards
room.image_source_model()
room.compute_rir()

[fakufaku]

Thanks for providing the code.
I would imagine this is indeed a computational bottleneck. The complexity of the image source method for arbitrary rooms is exponential in the order, i.e., proportional num_walls ** max_order.
I would suggest to use the hybrid simulation method with max_order limited to 2 or 3 and the rest done by ray tracing.

[whojavumusic]

True, but then the ray tracing does not support directivity so it wouldn’t make a realistic reverberation. Is there a plan to implement directivities to ray tracing anytime soon?

[fakufaku]

I see.
I am currently implementing directivities for ray tracing but there is no clear timeline for when that will be ready.
One thing you could try to do is to combine an RIR with low order using ISM with directivities with one using ray tracing but without the directivity for the tail.

[whojavumusic]

True, but that would defy the whole point of generating complex rooms if the later tail is isotropic. Anyway, Looking forward to the next update with RT with directivities included. Thanks :)