ESTIMATING SOUND SOURCE LOCATION USING PARTICLE FILTERING Russian patent published in 2014 - IPC G01S3/803 

FIELD: physics, acoustics.

SUBSTANCE: invention relates to estimation of the location of a sound source using particle filtering, particularly estimation of the location of a sound source for a multimodal audio-visual communication application. A sound source location is estimated by particle filtering where the particles represent a probability density function for a state variable comprising the sound source location. The method includes determining the weight coefficient for a particle in response to correlation between estimated acoustic transfer functions from the sound source to at least two sound recording positions. A weight coefficient update function may specifically be determined deterministically from the correlation and thus the correlation may be used as a pseudo-likelihood function for the measurement function of the particle filtering. The acoustic transfer functions may be determined from an audio beamforming towards the sound source. The audio weight coefficient may be combined with a video weight coefficient to generate a multimodal particle filtering approach.

EFFECT: high adaptability, easier estimation of the location of the sound source while increasing accuracy and improving efficiency.

15 cl, 9 dwg