Individualization of Head Related Transfer Functions Based on Radial Basis Function Neural Network

Lian, Meng; Wang, Xiaochen; Chen, Wei; Ai, Chunling; Hu, Ruimin

doi:10.1109/icme.2018.8486494

Cited by 5 publications

(3 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Most research on low-dimensional feature extraction of HRTF uses methods based on principal component analysis (PCA) [5,6,7] . However, such methods are difficult to effectively express the complex nonlinear relationships among low-dimensional features of HRTF [8] .…”

Section: Proposed Methodsmentioning

confidence: 99%

“…However, this also leads to retaining more low-dimensional data, thereby reducing the compression capability of the original data. Therefore, in the comparative experiments with the PCA method, we followed the algorithm in reference [7] to perform PCA analysis on the HRTF samples in the training set, obtaining principal component information that expresses 82%, 90%, and 95% of the overall variance, respectively. Subsequently, using the obtained principal component information, we examined the HRTF samples in the ARI dataset, analyzing the low-dimensional data quantity and highdimensional reconstruction accuracy achievable under different principal component conditions.…”

Section: Figure 4: Comparison Of the Reconstruction Effects Of Differ...mentioning

confidence: 99%

See 1 more Smart Citation

HRTF low-dimensional representation based on deep convolutional autoencoder and attention mechanism

2024

AJCIS

View full text Add to dashboard Cite

Head-Related Transfer Function (HRTF) depicts the reflection and scattering effects of the environment and the human body on sound during the transmission of sound signals from the sound source to the human ear and contains a large amount of auditory cue information for auditory localization. Due to the high-dimensional complexity and nonlinear nature of the sample data of HRTF itself, it creates difficulties in analyzing the relationship between the auditory localization cues of HRTF and the spatial orientation and morphological features of the human body. The traditional lowdimensional representation makes it difficult to effectively deal with the complex nonlinear relationships between multiple auditory cues in HRTF, resulting in performance degradation. To solve this problem, this study proposes a low-dimensional representation method for HRTF based on a deep convolutional autoencoder. The method considers that HRTF spectral features have the property of continuous variation in three-dimensional space and integrates the nonlinear relationships of full-space HRTF features by modeling the natural spatial attributes of the HRTF ensemble data. Firstly, the attention mechanism is introduced in the encoder, which solves the bias caused by mapping HRTF to a 3D tensor for convolution operation and mines the intrinsic features implied between the spatial orientations of HRTF neighborhoods and neighboring spectra, which improves the low-dimensional representation ability of the network. Secondly, the combination of dense connectivity and attention mechanism in the decoder according to the characteristics of different levels guarantees the effective delivery of lowdimensional features. Experimental results on several publicly available HRTF datasets show that the proposed model outperforms traditional methods in the low-dimensional representation and reconstruction of HRTFs and realizes high-performance low-dimensional representation and reconstruction of HRTFs.

show abstract

Section: Proposed Methodsmentioning

confidence: 99%

Section: Figure 4: Comparison Of the Reconstruction Effects Of Differ...mentioning

confidence: 99%

HRTF low-dimensional representation based on deep convolutional autoencoder and attention mechanism

2024

AJCIS

View full text Add to dashboard Cite

show abstract

“…He et al synthesized HRTFs using a sparse representation with different pre-processes and post-processes trained from anthropometric parameters [22]. Furthermore, the radial basis neural network has been utilized in HRTF estimation based on anthropometric parameters and achieved promising performance [23]. Qi et al proposed a sparsity-constrained weight mapping to individualize HRTFs, which obtained optimal weights to combine HRTFs based on the relationships of anthropometric parameters between the trained subjects and the target subject [24].…”

Section: Introductionmentioning

confidence: 99%

A Hybrid Algorithm for Predicting Median-Plane Head-Related Transfer Functions from Anthropometric Measurements

2019

View full text Add to dashboard Cite

Since head-related transfer functions (HRTFs) represent the interactions between sounds and physiological structures of listeners, anthropometric parameters represent a straightforward way to customize (or predict) individualized HRTFs. This paper proposes a hybrid algorithm for predicting median-plane individualized HRTFs using anthropometric parameters. The proposed hybrid algorithm consists of three parts: decomposition of HRTFs; selection of key anthropometric parameters; and establishing a prediction formula. Firstly, an independent component analysis (ICA) is applied to median-plane HRTFs from multiple subjects to obtain independent components and subject-dependent weight coefficients. Then, a factor analysis is used to select key anthropometric parameters relevant to HRTFs. Finally, a regression formula that connects ICA weight coefficients to key anthropometric parameters is established by a multiple linear regression. Further, the effectiveness of the proposed hybrid algorithm is verified by an objective evaluation via spectral distortion and a subjective localization experiment. The results show that, when compared with generic Knowles Electronics Manikin for Acoustic Research (KEMAR) HRTFs, the spectral characteristics of the predicted HRTFs are closer to those of the individualized HRTFs. Moreover, the predicted HRTFs can alleviate front-back and up-down confusion and improve the accuracy of localization for most subjects.

show abstract

HRTF Representation with Convolutional Auto-encoder

Chen

Wang

et al. 2019

MultiMedia Modeling

View full text Add to dashboard Cite

Individualization of Head Related Transfer Functions Based on Radial Basis Function Neural Network

Cited by 5 publications

References 17 publications

HRTF low-dimensional representation based on deep convolutional autoencoder and attention mechanism

HRTF low-dimensional representation based on deep convolutional autoencoder and attention mechanism

A Hybrid Algorithm for Predicting Median-Plane Head-Related Transfer Functions from Anthropometric Measurements

HRTF Representation with Convolutional Auto-encoder

Contact Info

Product

Resources

About