Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Head-Related Transfer Function Individualization Using Anthropometric Features and Spatially Independent Latent Representation

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read HRTF individualization can be reframed as a two-stage problem: predict a compact source-position-independent latent prototype from a listener's anthropometric measurements, then decode it with a pretrained autoencoder; on CIPIC and HUTUBS t

desk verdict A genuinely neat idea for pooling HRTF datasets with different grids, but the evaluation has a likely leakage flaw that needs to be addressed before the accuracy claim holds. read the letter →

arxiv 2508.16176 v1 pith:R27PVTOE submitted 2025-08-22 cs.SD eess.AS

classification cs.SDeess.AS
keywords HRTFindividualizationanthropometricparameterslatentrepresentationsource-positionindependenceautoencoderdiffusionmodeldatasetmerginglog-spectraldistortion
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proposes making individualized head-related transfer functions by first turning each listener's head, torso, and pinna measurements into a compact latent prototype that does not depend on sound-source direction, then letting a pretrained decoder expand that prototype into full magnitude spectra for every direction. This matters because HRTF datasets are small and use incompatible measurement grids, so direct network training is data-hungry and cannot pool datasets. The paper reports that on CIPIC and HUTUBS, the prototype DNN achieves lower log-spectral distortion than direct DNN baselines (e.g., 4.97 vs 4.99 dB on CIPIC) while using about 32k parameters instead of 82M, and that the latent-space diffusion model also generalizes better across datasets than an HRTF-space diffusion model.

What carries the argument

The central object is a source-position-conditioned and frequency-conditioned autoencoder whose encoder produces per-source-position latent codes that are averaged over source positions into a 'prototype.' The prototype is source-position-independent, so the decoder can be conditioned on any direction at synthesis time, and the architecture can train on datasets with incompatible measurement grids. The prototype is the bottleneck through which anthropometry is mapped to HRTF: a small DNN or a latent diffusion model learns to predict prototypes from anthropometric parameters, and the pretrained decoder turns them into HRTF magnitudes.

What would settle it

Retrain the autoencoder using only training-set subjects, explicitly excluding the held-out test subjects from pretraining, then rerun the prototype DNN evaluation on the same CIPIC and HUTUBS test sets. If the log-spectral distortion rises to or above the direct-DNN baselines of 4.99 and 5.09 dB, the reported advantage depends on decoder familiarity with test ears rather than on the source-position-independent latent representation.

Watch

Extended reading notes

Core claim

The central discovery is that HRTF personalization can be decomposed: a subject-specific latent 'prototype' independent of source position captures who the listener is, and direction/frequency structure is delegated to a conditioned autoencoder decoder. Once the decoder is pretrained, estimating an HRTF for a new listener reduces to predicting a low-dimensional prototype from 23 anthropometric parameters; the same frozen decoder then synthesizes all directions. On CIPIC and HUTUBS test sets, this two-stage scheme outperforms DNNs and diffusion models that estimate HRTF magnitudes directly, and it is the only scheme in the comparison that can be jointly trained on both datasets because the la

Load-bearing premise

The accuracy comparison assumes that pretraining the decoder on some of the same subjects later used for testing does not leak their ear characteristics; the paper adds subjects without anthropometric data for pretraining (Section 5.1) and never states that test subjects are excluded from that pretraining pool.

Editorial extensions

If this is right

  • Multiple HRTF datasets with different measured direction sets can be pooled in one training run, directly attacking the small-data problem in HRTF individualization.
  • Individualizing a new listener becomes a lightweight regression task: predict a 64-dimensional prototype per frequency from 23 anthropometric measurements instead of the full magnitude grid.
  • The direct 82M-parameter HRTF DNN can be replaced by a roughly 32k-parameter prototype DNN plus a shared decoder, with lower log-spectral distortion on the tested datasets.
  • Latent-space diffusion models inherit the same dataset-pooling benefit, whereas HRTF-space diffusion cannot be jointly trained across CIPIC and HUTUBS because of incompatible source positions.
  • Because the decoder is frozen after pretraining, the per-subject cost of deployment scales only with the prototype estimation network.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The reported accuracy margins over direct DNN baselines are small (roughly 0.02–0.12 dB); the larger practical win may be parameter efficiency and the ability to merge datasets, not dramatically lower spectral error.
  • If the paper's pretraining protocol does not exclude test subjects from the autoencoder's pretraining pool, as Section 5.1's wording leaves open, the frozen decoder may already be familiar with the test subjects' ears; retraining with strict subject exclusion would settle whether the accuracy gain survives.
  • A source-position-independent prototype could serve as a reusable subject embedding for other spatial-audio tasks, such as similarity-based HRTF retrieval, transfer to unmeasured directions, or uncertainty-aware probabilistic rendering when generated by a diffusion model.
  • The same two-stage decomposition may transfer to other low-data, high-dimensional personalization problems where global structure is shared across conditions but subject-specific variation is low-dimensional.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes a two-stage HRTF individualization method. First, a source-position- and frequency-conditioned autoencoder is pretrained to map HRTF log-magnitudes to source-position-independent latent vectors ('prototypes') by averaging encoded representations over source positions. Second, a lightweight DNN or a latent diffusion model (DM) is trained to predict these prototypes from anthropometric parameters; the frozen decoder then converts the predicted prototype into an HRTF magnitude. Experiments on CIPIC and HUTUBS compare the proposed prototype DNN/DM with direct HRTF DNN/DM baselines. The proposed prototype DNN reports the lowest LSD in all conditions (e.g., 4.97 vs 4.99 dB on CIPIC; 5.01 vs 5.09 dB on HUTUBS) while using far fewer trainable parameters.

Significance. The core idea of learning a source-position-independent latent space is original and practically relevant, as it enables combining HRTF datasets with different measured source positions without resampling. The manuscript also provides code, which is a strength. If the reported accuracy gains are valid, the method is a useful contribution to HRTF individualization because it drastically reduces the number of parameters and allows multi-dataset training. However, the experimental evidence as presented is not yet convincing: the evaluation protocol is ambiguous on a point that could invalidate the comparison, the parameter-count comparison is confounded, and the small reported differences are not supported by statistical tests.

major comments (3)
  1. [§5.1, Table 2] The evaluation protocol does not unambiguously state whether the held-out test subjects were excluded from autoencoder pretraining. The text says 'both the autoencoder and the individualization networks were trained on the same data subsets' but also says that 10 and 3 subjects 'were additionally included' for autoencoder pretraining. If the autoencoder's pretraining pool included the test subjects (e.g., because it was trained on the full dataset plus the extra subjects), the frozen decoder has already observed each test subject's HRTFs, giving the proposed method access to target information that the baselines do not have. Since the reported LSD differences are 0.02–0.08 dB, well within one standard deviation, leakage could fully explain the observed advantage. This is load-bearing for the central claim. The authors must state explicitly that test subjects were excluded from autoencode
  2. [§5.3, Table 2] The comparison is confounded by the large parameter mismatch (32k vs 82M for the DNN baselines). The baseline HRTF DNN is severely overparameterized relative to the training set size (30–85 subjects), so its performance may be limited by overfitting rather than by the absence of a latent-space intermediate. The proposed method's advantage could stem from the pretrained autoencoder's inductive bias rather than from the source-position-independent prototype itself. To support the claim that the proposed representation is the cause of improvement, the authors should include a baseline with a comparable number of trainable parameters, or a baseline using a linear (e.g., PCA) latent representation.
  3. [§5.3, Table 2] Only five (CIPIC) and six (HUTUBS) test subjects are used. The reported means and standard deviations overlap substantially (e.g., 4.97±0.25 vs 4.99±0.23 dB on CIPIC). Without paired significance testing (e.g., Wilcoxon signed-rank on per-subject LSD differences), the claim that the proposed networks 'consistently outperformed' the baselines is not statistically supported. Please report per-subject results or a significance test.
minor comments (5)
  1. [§1] Typo: 'easier to obtained' should be 'easier to obtain'.
  2. [Fig. 1 caption] The caption contains stray '🔥 / ❄' symbols, apparently a rendering artifact. Please remove them in the final version.
  3. [§5.2.1] The clamping of the predicted prototype to [−3, 3] at each timestep is ad hoc. Provide a sensitivity analysis or a justification for this range.
  4. [Fig. 3] Panel (d) shows the autoencoder reconstruction, but no quantitative LSD is given for this panel. Adding it would provide a useful reference for the ceiling of the proposed method.
  5. [Table 2] The 'N/A' entries for the HRTF DNN are unexplained in the table or its caption; the reason (incompatible source positions across datasets) appears only in §5.3. Move or explain the N/A entries.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the evaluation protocol is self-contained, though the data-split wording in Sec. 5.1 is ambiguous.

full rationale

I traced the derivation chain of the proposed HRTF individualization method. The method consists of (1) pretraining a source-position-conditioned autoencoder that maps HRTF magnitudes to position-independent prototypes and back, and (2) training a DNN or diffusion model to predict those prototypes from anthropometric parameters, then decoding with the frozen autoencoder decoder. The LSD evaluation in Sec. 5.3 is conducted on held-out subjects in the usual supervised-learning sense: the individualization networks are trained on the training split, and the test subjects' prototypes are not provided to the network during training. I found no equation or definition in which the predicted output is defined in terms of the target HRTF, nor any parameter fitted to a subset of data and then reported as a prediction of the same data. The only conceivable circularity would be if the autoencoder pretraining included the test subjects, making the frozen decoder a memorized reconstructor of the test HRTFs. The paper states: 'Importantly, both the autoencoder and the individualization networks were trained on the same data subsets' and that the autoencoder 'additionally included 10 and 3 subjects without complete anthropometric parameters for its pretraining.' Since the test subjects are by definition subjects with complete anthropometric parameters, the quoted wording does not support the claim that test subjects were in the autoencoder pretraining set; it is at most an ambiguity that would constitute a data-split correctness issue, not a circularity. The autoencoder architecture itself is adopted from prior work by overlapping authors ([18], [19]), but the paper does not use those citations as evidence for the central accuracy claim; instead, Table 2 and Fig. 3 report independent experimental comparisons against baselines. Therefore I find no self-definitional, fitted-input-called-prediction, self-citation-load-bearing, or other circular step.

Assumptions & free parameters 4 free parameters · 4 assumptions · 1 invented entities

The method rests on a learned latent bottleneck, the sufficiency of 23 anthropometric measurements, and the validity of pooling two datasets with different measurement geometries. No physical constants or machine-checked derivations are involved; all quantitative claims are empirical and depend on the hyperparameters listed above.

free parameters (4)
  • Latent dimension D = 64
    Selected by the authors as the autoencoder bottleneck dimension; controls how much HRTF detail the prototype can carry.
  • FFM frequency count K = 8 for autoencoder, 16 for individualization networks
    Hand-chosen hyperparameters of the Fourier feature mapping; K directly affects the conditioning of source position and frequency.
  • DDIM inference parameters (w, eta) = w=4, eta=0.2 (prototype DM); w=2, eta=0.25 (HRTF DM)
    Empirically tuned in Section 5.2.3; these values control guidance strength and stochasticity of the generated latents/HRTFs.
  • Prototype clamping range = [-3, 3]
    Ad hoc clipping applied at each sampling timestep 'to stabilize generation' (Section 5.2.3).
assumptions (4)
  • domain assumption HRTF phase can be reconstructed from magnitude via minimum phase plus separately estimated ITD
    Section 2: the method estimates only log-magnitude, which is standard in the field but is a simplifying assumption about the sufficiency of magnitude information.
  • domain assumption The 23 shared anthropometric parameters between CIPIC and HUTUBS are sufficient to determine the subject-specific part of the HRTF
    Section 2 lists the 23 parameters; the method implicitly assumes these quantities capture the anthropometric variation that drives HRTF differences.
  • ad hoc to paper Averaging the position-conditioned encoder outputs over source positions yields a source-position-independent prototype that retains subject identity
    Section 4.1 defines the prototype as a mean over source positions; no proof is given that this average preserves the individual-specific information needed for reconstruction.
  • domain assumption CIPIC (1.0 m) and HUTUBS (1.47 m) can be combined in one autoencoder despite different measurement distances
    Section 5.1 merges the datasets without distance normalization; the method relies on the network absorbing the distance difference into the position conditioning.
invented entities (1)
  • Source-position-independent prototype latent vector
    purpose: A compact subject-specific code, averaged over sound-source positions, which the individualization network predicts from anthropometry and the decoder converts into HRTF magnitude.
    The prototype has no predictive handle outside the fitted autoencoder; its usefulness is measured only by internal LSD values within the same framework.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Head-Related Transfer Function Individualization Using Anthropometric Features and Spatially Independent Latent Representation." pith.science (2026). https://pith.science/paper/R27PVTOE

@misc{pith2026250816176,
  author       = {Pith},
  title        = {Pith review of: Head-Related Transfer Function Individualization Using Anthropometric Features and Spatially Independent Latent Representation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/R27PVTOE}},
  note         = {Machine review of arXiv:2508.16176}
}
read the original abstract

A method for head-related transfer function (HRTF) individualization from the subject's anthropometric parameters is proposed. Due to the high cost of measurement, the number of subjects included in many HRTF datasets is limited, and the number of those that include anthropometric parameters is even smaller. Therefore, HRTF individualization based on deep neural networks (DNNs) is a challenging task. We propose a HRTF individualization method using the latent representation of HRTF magnitude obtained through an autoencoder conditioned on sound source positions, which makes it possible to combine multiple HRTF datasets with different measured source positions, and makes the network training tractable by reducing the number of parameters to be estimated from anthropometric parameters. Experimental evaluation shows that high estimation accuracy is achieved by the proposed method, compared to current DNN-based methods.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. HRTFformer: A Spatially-Aware Transformer for Individual HRTF Upsampling in Immersive Audio Rendering

    cs.SD 2025-10 conditional novelty 5.0 of 10

    HRTFformer reconstructs high-resolution head-related transfer functions from as few as three measured directions using a transformer in the spherical harmonic domain, beating prior methods in accuracy.

Reference graph

Works this paper leans on

39 extracted references · 38 canonical work pages · cited by 1 Pith paper

  1. [14]

    Autoencoding HRTFS for DNN based HRTF personalization using anthropometric features,

    T.-Y . Chen, T.-H. Kuo, and T.-S. Chi, “Autoencoding HRTFS for DNN based HRTF personalization using anthropometric features,” in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP) , 2019, pp. 271–275

  2. [17]

    A hybrid approach to structural modeling of individualized HRTFs,

    R. Miccini and S. Spagnol, “A hybrid approach to structural modeling of individualized HRTFs,” in Proc. IEEE Conf. Virtual Reality and 3D User Interfaces Abstracts and Workshops (VRW) , 2021, pp. 80–85

  3. [1]

    Blauert, Spatial hearing: the psychophysics of human sound localization

    J. Blauert, Spatial hearing: the psychophysics of human sound localization . MIT press, 1997

  4. [2]

    Localization using nonindividualized head-related transfer functions,

    E. M. Wenzel, M. Arruda, D. J. Kistler, and F. L. Wightman, “Localization using nonindividualized head-related transfer functions,” J. Acoust. Soc. Amer ., vol. 94, no. 1, pp. 111–123, 1993

  5. [3]

    Binaural technique: Do we need individual recordings?

    H. Møller, M. F. Sørensen, C. B. Jensen, and D. Hammershøi, “Binaural technique: Do we need individual recordings?” J. Audio Eng. Soc., vol. 44, pp. 451–469, jun 1996

  6. [4]

    HRTF personaliza- tion using anthropometric measurements,

    D. Zotkin, J. Hwang, R. Duraiswami, and L. Davis, “HRTF personaliza- tion using anthropometric measurements,” in Proc. IEEE Int. Workshop Appl. Signal Process. Audio Acoust. (WASPAA) , 2003, pp. 157–160

  7. [5]

    A survey on machine learning techniques for head-related transfer function individualization,

    D. Fantini, M. Geronazzo, F. Avanzini, and S. Ntalampiras, “A survey on machine learning techniques for head-related transfer function individualization,” IEEE Open J. Signal Process. , vol. 6, pp. 30–56, 2025

  8. [6]

    A wide dataset of ear shapes and pinna- related transfer functions generated by random ear drawings,

    C. Guezenoc and R. Seguier, “A wide dataset of ear shapes and pinna- related transfer functions generated by random ear drawings,” J. Acoust. Soc. Amer ., vol. 147, no. 6, pp. 4087–4096, 2020

Show all 39 references
  1. [7]

    Prediction of head related transfer functions using machine learning approaches,

    R. Fernandez Martinez, P. Jimbert, E. M. Sumner, M. Riedel, and R. Unnthorsson, “Prediction of head related transfer functions using machine learning approaches,” Acoustics, MDPI , vol. 5, no. 1, pp. 254– 267, 2023

  2. [8]

    Global HRTF personal- ization using anthropometric measures,

    Y . Wang, Y . Zhang, Z. Duan, and M. Bocko, “Global HRTF personal- ization using anthropometric measures,” Proc. AES Conv. , no. 10502, august 2020

  3. [9]

    Magnitude modelling of individualized HRTFs using dnn based spherical harmonic analysis,

    J. Xi, W. Zhang, and T. D. Abhayapala, “Magnitude modelling of individualized HRTFs using dnn based spherical harmonic analysis,” in Proc. IEEE Int. Workshop Appl. Signal Process. Audio Acoust. (WASPAA) , 2021, pp. 266–270

  4. [10]

    HRTF personalization based on artificial neural network in individual virtual auditory space,

    H. Hu, L. Zhou, H. Ma, and Z. Wu, “HRTF personalization based on artificial neural network in individual virtual auditory space,” vol. 69, no. 2, pp. 163–172, 2008

  5. [11]

    Individualization of head related transfer functions based on radial basis function neural network,

    L. Meng, X. Wang, W. Chen, C. Ai, and R. Hu, “Individualization of head related transfer functions based on radial basis function neural network,” in Proc. IEEE International Conference on Multimedia and Expo (ICME) , 2018, pp. 1–6

  6. [12]

    Modeling of individual HRTFs based on spatial principal component analysis,

    M. Zhang, Z. Ge, T. Liu, X. Wu, and T. Qu, “Modeling of individual HRTFs based on spatial principal component analysis,” IEEE/ACM Trans. Audio, Speech, Lang. Process. , vol. 28, pp. 785–797, 2020

  7. [13]

    Anthropometric- based customization of head-related transfer functions using isomap in the horizontal plane,

    F. Grijalva, L. Martini, S. Goldenstein, and D. Florencio, “Anthropometric- based customization of head-related transfer functions using isomap in the horizontal plane,” in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP), 2014, pp. 4473–4477

  8. [15]

    Pre-trained-based individualization model for real-time spatial audio rendering system,

    J. Lu and X. Qi, “Pre-trained-based individualization model for real-time spatial audio rendering system,” IEEE Access, vol. 9, pp. 128 722–128 733, 2021

  9. [16]

    An individualization approach for head-related transfer function in arbitrary directions based on deep learning,

    D. Yao, J. Zhao, L. Cheng, J. Li, X. Li, X. Guo, and Y . Yan, “An individualization approach for head-related transfer function in arbitrary directions based on deep learning,” JASA Express Letters , vol. 2, no. 6, p. 064401, 06 2022

  10. [18]

    Head-related transfer function interpolation from spatially sparse measurements using autoencoder with source position conditioning,

    Y . Ito, T. Nakamura, S. Koyama, and H. Saruwatari, “Head-related transfer function interpolation from spatially sparse measurements using autoencoder with source position conditioning,” in Proc. Int. Workshop Acoust. Signal Enhancement (IWAENC) , Sep. 2022

  11. [19]

    Spatial upsampling of head-related transfer function using neural network conditioned on source position and frequency,

    Y . Ito, T. Nakamura, S. Koyama, S. Sakamoto, and H. Saruwatari, “Spatial upsampling of head-related transfer function using neural network conditioned on source position and frequency,” IEEE Open J. Signal Process., 2025, (in press)

  12. [20]

    The CIPIC HRTF database,

    V . R. Algazi, R. O. Duda, D. M. Thompson, and C. Avendano, “The CIPIC HRTF database,” inProc. IEEE Int. Workshop Appl. Signal Process. Audio Acoust. (WASPAA) . IEEE, 2001, pp. 99–102

  13. [21]

    Brinkmann, D

    F. Brinkmann, D. Manoj, R. Pelzer, J. J. Wohlgemuth, F. Seipel, D. V oss, P. Grosche, and S. Weinzierl. The HUTUBS head-related transfer function (HRTF) database. Doi: 10.14279/depositonce-8487, (accessed May 6, 2022). [Online]. Available: http://dx.doi.org/10.14279/depositonce-8487

  14. [22]

    Computed hrirs and ears database for acoustic research,

    S. Ghorbal, X. Bonjour, and R. S ´eguier, “Computed hrirs and ears database for acoustic research,” in Proc. 148th AES Conv. Audio Engineering Society, 2020

  15. [23]

    Estimation of HRTFs on the horizontal plane using physical features,

    T. Nishino, N. Inoue, K. Takeda, and F. Itakura, “Estimation of HRTFs on the horizontal plane using physical features,” Applied Acoustics , vol. 68, no. 8, pp. 897–908, 2007, head- Related Transfer Function and its Applications

  16. [24]

    A multilinear (tensor) framework for HRTF analysis and synthesis,

    G. Grindlay and M. A. O. Vasilescu, “A multilinear (tensor) framework for HRTF analysis and synthesis,” in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP) , vol. 1, 2007, pp. I–161–I–164

  17. [25]

    HRTF magnitude synthesis via sparse representation of anthropometric features,

    P. Bilinski, J. Ahrens, M. R. P. Thomas, I. J. Tashev, and J. C. Platt, “HRTF magnitude synthesis via sparse representation of anthropometric features,” in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP), 2014, pp. 4468–4472

  18. [26]

    HRTF personalization based on weighted sparse representation of anthropometric features,

    M. Zhu, M. Shahnawaz, S. Tubaro, and A. Sarti, “HRTF personalization based on weighted sparse representation of anthropometric features,” in Proc. Int. Conf. 3D Immersion (IC3D) , 2017, pp. 1–7

  19. [27]

    Generation of the amplitude spectra of the individual head-related transfer functions in the upper median plane based on the anthropometry of the listener’s pinnae,

    K. Iida, H. Shimazaki, and M. Oota, “Generation of the amplitude spectra of the individual head-related transfer functions in the upper median plane based on the anthropometry of the listener’s pinnae,” Applied Acoustics , vol. 155, pp. 280–285, 2019

  20. [28]

    Advancement of individualized head- related transfer functions (HRTFs) in perceiving the spatialization cues: Case study for an integrated HRTF individualization method,

    L. Wang, X. Zeng, and X. Ma, “Advancement of individualized head- related transfer functions (HRTFs) in perceiving the spatialization cues: Case study for an integrated HRTF individualization method,” Applied Sciences, vol. 9, no. 9, 2019

  21. [29]

    Modelling individual head-related transfer function (HRTF) based on anthropometric parameters and generic HRTF amplitudes,

    R. Zhang, R. Meng, J. Sang, Y . Hu, X. Li, and C. Zheng, “Modelling individual head-related transfer function (HRTF) based on anthropometric parameters and generic HRTF amplitudes,” CAAI Trans. Intell. Technol. , vol. 8, no. 2, pp. 364–378, 2023

  22. [30]

    Towards HRTF personalization using denoising diffusion models,

    J. C. Albarrac ´ın S ´anchez, L. Comanducci, M. Pezzoli, and F. Antonacci, “Towards HRTF personalization using denoising diffusion models,” in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP) , 2025, pp. 1–5

  23. [31]

    Mish: A self regularized non-monotonic neural activation function,

    D. Misra, “Mish: A self regularized non-monotonic neural activation function,” in Proc. Brit. Mach. Vis. Conf. , 2020

  24. [32]

    HyperNetworks,

    D. Ha, A. M. Dai, and Q. V . Le, “HyperNetworks,” in Proc. Int. Conf. Learn. Repr ., 2017

  25. [33]

    Fourier features let networks learn high frequency functions in low dimensional domains,

    M. Tancik, P. Srinivasan, B. Mildenhall, S. Fridovich-Keil, N. Raghavan, U. Singhal, R. Ramamoorthi, J. Barron, and R. Ng, “Fourier features let networks learn high frequency functions in low dimensional domains,” Adv. Neural. Inf. Process. Syst. , vol. 33, pp. 7537–7547, 2020

  26. [34]

    High-resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High-resolution image synthesis with latent diffusion models,” in Proc. IEEE/CVF Conf. Comput. Vision Pattern Recog. (CVPR) , 2022, pp. 10 684–10 695

  27. [35]

    Diffusion models: A comprehensive survey of methods and applications,

    L. Yang, Z. Zhang, Y . Song, S. Hong, R. Xu, Y . Zhao, W. Zhang, B. Cui, and M.-H. Yang, “Diffusion models: A comprehensive survey of methods and applications,” ACM Comput. Surveys , vol. 56, no. 4, pp. 1–39, 2023

  28. [36]

    Denoising diffusion implicit models,

    J. Song, C. Meng, and S. Ermon, “Denoising diffusion implicit models,” in Proc. Int. Conf. Learn. Repr . , 2021

  29. [37]

    Classifier-free diffusion guidance,

    J. Ho and T. Salimans, “Classifier-free diffusion guidance,” 2022. [Online]. Available: https://arxiv.org/abs/2207.12598

  30. [38]

    Understanding and improving layer normalization,

    J. Xu, X. Sun, Z. Zhang, G. Zhao, and J. Lin, “Understanding and improving layer normalization,” in Proc. Neural Information Processing Systems (NeurIPS) , 2019

  31. [39]

    Decoupled weight decay regularization,

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” in Proc. Int. Conf. Learn. Repr . , 2019

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.