Abstract:
An improvement to our previous DOA (direction of arrival) estimation method with a two-microphone array system is proposed in this paper. The method utilizes the harmonic structure in vowel sounds to generate multichannel frequency-domain array data. The key concept is to introduce a selection mechanism of harmonics according their SNR values, in order to realize robustness for both vowel and speaker feature differences. The simulation results indicate that the estimation accuracy is improved by introducing the new algorithm.