Paper Summary
Click the triangle to expand/collapse the contents.
Demo Sound Samples
Here are sound samples that are rendered using the HRTFs that are simulated (FABIAN), estimated (Ours), or linearly interpolated (baseline).
- GT-2: Rendered using the simulated HRTF of FABIAN ($2\degree$interval HRTFs)
- Ours-2: Rendered using the interpolated HRTF with the proposed method (estimated $2\degree$interval HRTFs from the neighborhood $q_i\in\mathbf{n}_p$ with $2\degree$interval)
- Ours-12: Rendered using the spatially upsampled (super-resolution) HRTF with the proposed method (recovered $2\degree$interval HRTFs from the neighborhood $q_i\in\mathbf{n}_p$ with $12\degree$interval)
- Baseline-12: Rendered using the linearly interpolated HRTF (interpolated $2\degree$interval HRTFs from $12\degree$interval)
- Baseline-40: Rendered using the linearly interpolated HRTF (interpolated $2\degree$interval HRTFs from $40\degree$interval)
- Baseline-90: Rendered using the linearly interpolated HRTF (interpolated $2\degree$interval HRTFs from $90\degree$interval)
<aside>
☝ Reminder: Our model was trained using HUTUBS database (which uses a different coordinate system from FABIAN; see here).
</aside>
We simulate the binaural audio from three types of monaural samples: pink noise, white noise, and human speech.
<aside>
💡 Be sure to listen to the samples with earphones!
Listening with headphones or speakers does not accurately reflect the HRTF rendered in the audio.
</aside>
<aside>
💡 Listen to the speech sample below, and adjust the volume on your device to the appropriate level.
bonafide-2.wav
</aside>
Frontal Plane Simulation
Horizontal Plane Simulation
Median Plane Simulation
<aside>
👂 Pay close attention to how well the location of the sound you hear matches the path of each plane (red circles in the figure above).
</aside>