Spatial grouping as a method to improve personalized head-related transfer function prediction.

Area of Science:

• Acoustics and Audio Engineering
• Machine Learning and Artificial Intelligence

Background:

• Head-related transfer functions (HRTFs) are crucial for spatial audio perception.
• Accurate HRTF estimation is computationally intensive, especially with neural network models.
• Existing angle-specific models offer high performance but require significant computational resources.

Purpose of the Study:

• To develop a computationally efficient method for HRTF estimation.
• To improve the performance of neural network-based HRTF prediction models.
• To balance performance and computational cost in HRTF estimation.

Main Methods:

• Proposing a novel method involving spatial grouping of HRTF data into subspaces.
• Reducing variance within each subspace through targeted data grouping.
• Training individual HRTF predicting neural networks for each subspace.

Main Results:

• The proposed method demonstrates superior performance compared to global models.
• The spatial grouping approach outperforms traditional angle-specific models.
• Different grouping strategies show effectiveness for ipsilateral and contralateral sides.

Conclusions:

• Spatially grouping HRTF data offers an effective strategy for improving prediction accuracy.
• The proposed method provides a balanced solution for HRTF estimation, optimizing performance and computational load.
• This technique advances the field of personalized spatial audio and virtual acoustics.