Abstract
Human speeches are sound signals intermittently repeating sound and silence sections. When a traditionally well-known DOA (direction-of-arrival) estimation approach such as MUSIC and ESPRIT is applied to estimate the DOA of human speech, the resulting estimates are degraded by receiving the influence of noises and reflecting waves superimposed in the silence sections. From this point of view, the present paper proposes a framewise DOA estimation of a target sound source with two microphones taking advantage of the sparsity of speech sounds. Several experiments were carried out to verify the proposed DOA estimation. It has been found that the proposed estimation is valid for the true DOA being －30◦ to 30◦ under the condition where SNR≥ 15[dB] and RT60 ≤ 200[msec].
