Narrow Results

The method of estimating Head-Related Impulse Responses (HRIRs) from impulse responses involving the reflection and/or noise was investigated. This method is based on the Auto Regressive (AR) model of HRIR. The AR coefficients are estimated using the conventional Linear Prediction (LP) algorithm. The data used for estimation are the part of the response which is regarded as the direct component of the impulse response from the sound source to the ear, i.e., the part of the HRIR. A computer simulation in which the method is applied to the estimation of the HRIRs of the Head-and-Torso Simulator (HATS) showed the following: (1) SDR was improved in some directions examined when the order of AR coefficients is half of the cutout points. Otherwise the improvement of SDR was not observed. Although the reason for such occasional improvement remains unclear, the improvement demonstrates that the proposed method can better estimate the part of the HRIRs lost by reflection and/or noise; (2) The number of samples used for the computation of AR coefficients greatly affects the estimation accuracy. Ideally, the whole waveform of the HRIR is useful for estimation of the AR coefficients, which shows that the proposed method brings about accurate estimation in an ideal situation.