The sound source localization based on Gaussian mixture models (GMMs) in a neck-band microphone array module is investigated for the localization of the alarming sound. The sound signals are measured by 6 MEMS microphones mounted on the neck-band placed on a mannequin, and the sound source location such as an azimuthal angle or a distance from the sound source is analysed by steered response power with the phase transform (SRP-PHAT) or the GMM classifier using the inter-aural time differences (ITDs) between the microphone pairs and the ratio between direct and reverberant signal components (DRR) which is estimated by measuring cross-correlation ratio (CCR) or coherent-to-diffuse ratio (CDR). The accuracies of the azimuthal angle and the distance from the sound source are comparatively evaluated and analysed between GMM and SRP-PHAT methods. We found that GMM method was better than SRP-PHAT, and the distance as well as the azimuthal angle was well estimated. This technology can be applied to the safety assistance service for disabled people who cannot hear environmental alarming sounds.