We should use the inverse distance clamped model (the default), but we first need to handle muting sounds that are beyond their max distance. Linear attenuation doesn't give a proper rolloff, but it makes the sounds silent at max distance.