The sound from speech when many people are gathered for a social event or in an eating establishment is a challenge for room acoustical modeling, because the sound power of the source varies with the level of the ambient noise, the so-called Lombard effect. A new method has been developed for modeling the ambient noise of many people speaking in a closed space. The sound source is a transparent surface placed just above the heads of people and radiating in all directions from a large number of randomly distributed points on the surface. The receivers are represented by a grid of points covering the same area as the source, but at the ears’ height. The sound power and spectrum of the source are based on the ANSI 3.5 standard and adjusted taking the Lombard effect into account. Only two parameters are involved: the number of people speaking and the calculated transfer function from the surface source to the grid of receivers; the latter is strongly dependent on the reverberation time in the room. The method has been tested and verified by comparison to measured data in three cases with very different reverberation times and with 380, 480 and 530 people, respectively. Auralization can be made to demonstrate the difficulty of conversation in ambient noise.