Localization of sound sources in rooms using an improved version of steered response power algorithm.

Abstract

Localization of sound sources inside a room is a challenging problem. The possible applications involving speech‐based source localization systems range from teleconferencing to home automation systems. For example, the localization of a speaker inside a conference room can be very useful to place the speaker in a remote room by means of a spatial audio reproduction system. Also, new applications will appear in home automation if accurate source localization systems are available in the future. One of the most robust approaches to source localization is the SRP‐PHAT algorithm, which has shown to provide very good localization results inside rooms with moderate reverberation. However, the computational cost is highly dependent on the spatial sampling and the number of microphones, making very difficult the localization of sound events if a coarse spatial sampling is used. In this paper, we propose an improvement of this method where sound events are not missed even if a very coarse grid is used. The improvement is based on a previous calculation of the existent cross‐correlation lags between spatially adjacent points in the grid, assuring that the non‐sampled space is covered in terms of cross‐correlation lags between microphone pairs when running the algorithm. Several experiments conducted in different rooms with complex acoustics confirm the validity and benefits of the proposed method.