Speaker
Description
The exponential sine sweep (ESS) technique became the common approach for measuring head-related transfer functions (HRTFs) since it allows separating the loudspeaker’s distortion from the HRTF by temporal windowing. This, however, fails with overlapping excitation signals. This work assesses the signal-to-noise ratio (SNR) benefit from nonlinear harmonic prediction and subtraction. The measured response is modeled as a superposition of a linear response, nonlinear harmonic distortions, and additive noise. Each harmonic distortion component is represented using a finite impulse response kernel. The kernel is estimated from a reference measurement. For evaluation, broadband and Bark-band limited SNR is computed. Preliminary results for frontal HRTF measurements show broadband SNR values of 30-45 dB depending on the loudspeaker direction, with the proposed harmonic subtraction improving these values by up to about 3 dB for the most affected (most elevated) channel. In perceptually relevant mid-to-high frequency regions, where spectral notches make the HRTF sensitive to measurement noise, mean SNR gains of 1-2 dB are observed. The results demonstrate that the proposed method has potential to reduce non-linear distortion and provide cleaner impulse responses and more accurate noise estimates in overlapping ESS-based HRTF measurements.