SOUND-SOURCE SEPARATION METHOD, DEVICE, AND PROGRAM

摘要

Provided are a sound-source separation method, device, and program in which, using close microphones, without requiring special analysis, sounds coming from a given direction can be either accentuated or suppressed with a low computational burden. A filter that includes a delay of a specified duration is applied to one of a pair of input signals inputted via microphones (L, R). After the application of said filter, a swap circuit (2) swaps every other sample of the pair of input signals (InR and InL), generating a pair of swapped signals (InA and InB). A coefficient-update circuit (3) updates a coefficient (m) on a per-sample basis by multiplying one of the swapped signals (InB) by said coefficient (m), then generating error signals for the swapped signals (InA and InB), and using said error signals to compute a recurrence formula for the coefficient (m). The input signals (InL and InR) are multiplied by the successively updated coefficients (m) and the results are outputted.