Pith. sign in

REVIEW 4 major objections 6 minor 82 references

ArrayDPS: Unsupervised Blind Speech Separation with a Diffusion Prior

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read ArrayDPS claims that blind multi-microphone speech separation can be done with only a pretrained single-speaker diffusion prior and the mixtures, and reports SDR above all unsupervised baselines, matching supervised models on ad-hoc arrays.

desk verdict ArrayDPS is a genuinely new and well-tested unsupervised multi-channel separation method, but its headline theoretical justification (Theorem B.1) does not hold as stated; the empirical case survives. read the letter →

arxiv 2505.05657 v3 pith:7AQRWSXT submitted 2025-05-08 eess.AS cs.LGcs.MMcs.SDeess.SP

Zhongweiyang Xu , Xulin Fan , Zhong-Qiu Wang , Xilin Jiang , Romit Roy Choudhury This is my paper · ORCID
classification eess.AScs.LGcs.MMcs.SDeess.SP
keywords blindspeechseparationdiffusionposteriorsamplingunsupervisedsourcearray-agnosticmicrophonearrayrelativeroomimpulseresponseforwardconvolutivepredictionindependentvectoranalysismulti-channelaudio
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

ArrayDPS claims that blind multi-microphone speech separation can be solved with no knowledge of the array geometry, the room acoustics, or the sources themselves: only a pretrained single-speaker diffusion prior and the recorded mixtures are needed. It reformulates separation as diffusion posterior sampling and, at each sampling step, replaces the intractable likelihood by a maximum-likelihood estimate of the relative room impulse responses computed with Forward Convolutive Prediction. On the 3-channel SMS-WSJ fixed-array test it reports 15.8 dB SDR, above the unsupervised UNSSOR baseline (15.4 dB) and close to the supervised TF-GridNet range (16.2 to 16.8 dB); on ad-hoc 4-channel Spatialized WSJ0-2Mix it reports 16.1 dB SDR, ahead of UNSSOR (15.2 dB) and of the supervised array-agnostic TF-GridNet (15.8 dB). If these numbers hold, an array-agnostic separator can be built without paired training data, and the same likelihood-approximation loop may apply to other multi-channel inverse problems.

What carries the argument

The load-bearing object is the virtual-source relative-RIR model: every speaker is a single virtual waveform, and each microphone channel is a filter-and-sum of those waveforms through unknown filters $g^k_{0\to c}(t)$, so the full mixing map $A$ is an operator on sources and filters rather than a known array response. ArrayDPS runs an EDM-style probabilistic flow ODE for posterior sampling and, inside each step, a maximum-likelihood filter-estimation subroutine: Forward Convolutive Prediction solves a weighted least-squares problem in the STFT domain with an analytical solution, which the paper proves is the ML estimator of the relative RIRs. The denoise-estimate-filter-backprop loop is fully differentiable, so the likelihood gradient can be computed without ever knowing the microphone positions or the room impulse responses.

What would settle it

Run ArrayDPS on simulated two-speaker mixtures with sensor noise swept from 30 dB down to 0 dB while keeping the room, T60, and source positions fixed; if the central claim is right, separation SDR should fall smoothly with the noise floor, whereas if the delta-RIR likelihood approximation fails, the SDR should collapse far faster than the mixture SNR predicts, especially without IVA initialization.

Watch

Extended reading notes

Core claim

The paper's central claim is that the separation posterior $p(s_{1:K}\mid x_{1:C})$ can be sampled even though the mixing operator is unknown, by estimating that operator inside every diffusion step. Each speaker is represented by a virtual source $s_k$, and each microphone observation is modeled as the sum over speakers of the convolution of that virtual source with an unknown relative RIR $g^k_{0\to c}$; the filters are re-estimated by maximum likelihood at every step using Forward Convolutive Prediction, and the residual between the reconstructed and recorded mixtures is back-propagated through the differentiable chain to form a likelihood score. The authors prove that the FCP estimate is the maximum-likelihood relative-RIR estimator, and they show the likelihood approximation is exact only when the conditional law of the relative RIRs is a delta function at that estimate. With IVA initialization and a reference-channel guidance term, the method reports SDR scores that beat all unsupervised baselines and, on ad-hoc arrays, match or exceed the supervised TF-GridNet baseline.

Load-bearing premise

The whole likelihood gradient rests on treating the FCP-estimated relative RIRs as if they were the true filters: the approximation is exact only when the filter posterior is a delta function, yet the evaluation mixtures are simulated with 20 to 30 dB sensor noise, so the approximation is assumed rather than proven to hold under that noise.

Editorial extensions

If this is right

  • On 3-channel SMS-WSJ, default ArrayDPS reaches 15.8 dB SDR without mixing supervision, beating UNSSOR (15.4 dB) and coming within about 0.5-1 dB of supervised TF-GridNet.
  • On ad-hoc 4-channel Spatialized WSJ0-2Mix, where microphone geometries vary per sample, ArrayDPS reaches 16.1 dB SDR, surpassing both the best unsupervised baseline (15.2 dB) and the supervised array-agnostic TF-GridNet (15.8 dB).
  • Selecting the maximum-likelihood sample from five draws raises the fixed-array SDR to 16.9 dB and the ad-hoc SDR to 16.6 dB, both above the comparable supervised models.
  • The method stays unsupervised and array-agnostic in underdetermined and three-speaker settings: it separates two speakers from two microphones where UNSSOR collapses, and on Spatialized WSJ0-3Mix it reports 12.8 dB SDR versus 10.6 dB for the supervised baseline.

Reading between the lines

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

  • The operator-estimation loop - fit the unknown filter by maximum likelihood, then backpropagate the reconstruction error - is a general recipe that could be applied to other blind array problems such as dereverberation, source localization, or calibration whenever a generative prior over the target signal exists.
  • Because the delta-RIR approximation is exact only at zero measurement noise, performance in noisier real rooms may degrade; a natural extension is to place a proper posterior over the relative RIRs instead of a point estimate, or to anneal the likelihood weight with the estimated noise level.
  • The observed dereverberating behavior of the virtual-source outputs suggests the anechoic diffusion prior is also acting as a regularizer; a testable variant would condition the prior on estimated room acoustics to let users trade off between separation fidelity and dereverberation.
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

4 major / 6 minor

Summary. ArrayDPS proposes an unsupervised, array-agnostic generative method for multichannel blind speech separation. It uses a pre-trained single-speaker diffusion prior and, at each diffusion step, estimates relative room impulse responses from the current noisy source estimates via Forward Convolutive Prediction (FCP); the resulting differentiable mixture-reconstruction loss is used as an approximate likelihood score. Independent Vector Analysis is used for initialization and early guidance, and reference-channel guidance is added as an empirical stabilizer. Experiments on SMS-WSJ and Spatialized WSJ0-2Mix/3Mix report SDR, SI-SDR, PESQ, and eSTOI gains over unsupervised baselines, competitiveness with a supervised TF-GridNet, and open-sourced code and demos.

Significance. If the theoretical derivation were solid, this would be a notable step: applying diffusion posterior sampling to a blind multi-channel array inverse problem, with array-agnostic and unsupervised operation and strong empirical results. The paper's concrete strengths are its extensive evaluation across 2/3/4/6 channels and 3 speakers, ablations, hyperparameter sensitivity studies, standard deviations over repeated samples, and released code and audio demos. However, the central likelihood-score derivation rests on an unproven decoupling claim in Theorem B.1 and an explicitly acknowledged delta-posterior approximation that the evaluated 20-30 dB sensor noise regime violates. The contribution is therefore currently more empirical than principled; a revision that corrects or honestly reframes the derivation, with supporting experiments, would make the work substantially stronger.

major comments (4)
  1. [Appendix B.2, Theorem B.1 (Eqs. 44-45)] The asserted equivalence between the joint maximum-likelihood filter objective and the per-source FCP objectives does not follow. Minimizing ||X_c - sum_k G_k * S_k||^2 over all filters couples the G_k through the empirical cross-correlations between S_j and S_k, and the claim that independence of the source distributions makes those cross-correlations zero is not valid for a given finite STFT realization; speech frames are neither independent nor W-disjoint orthogonal. Since Eq. 38 and hence the likelihood approximation in Eq. 34 depend on this equivalence, the central derivation is not established. The authors should either supply a proof with explicit assumptions (e.g., oracle orthogonal sources) or weaken the claim and validate the per-source FCP step empirically.
  2. [Appendix B.2, Eq. 48 and Appendix E.1] The approximation p(X_c | S_hat) ≈ p(X_c | S_hat, G_hat) is exact only when the posterior over relative RIRs is a delta at the FCP estimate. The paper acknowledges that this holds only with no measurement noise, while the evaluation mixtures are simulated with 20-30 dB sensor noise (Appendix E.1). This is a load-bearing gap because Eq. 34 is the mechanism that conditions the diffusion prior on the multichannel mixtures. The authors should quantify the bias of this approximation, for example by comparing likelihood gradients computed from oracle-relative-RIRs against those from FCP estimates, or explicitly treat Eq. 18 as a heuristic whose validity is demonstrated only through experiments.
  3. [Eq. 14 vs. Theorem B.1] Theorem B.1 assumes the FCP weight is lambda_kc(l,f) = 1/(2 sigma_N^2), but the implemented weight in Eq. 14 is a data-dependent mixture-energy normalization. Consequently, even the per-source FCP estimator is not exactly maximum likelihood under the paper's own noise model. The claim in Section 3.2 that 'FCP is equivalent to the maximum likelihood relative RIR estimator' should be qualified, or Theorem B.1 should be extended to the actual weighting.
  4. [Table 1, rows 2d-2e] The ablations show that IVA initialization and reference-channel guidance are decisive: without IVA initialization, default ArrayDPS drops to 8.5 dB SDR with high standard deviation, and without both IVA and reference guidance it drops to 0.9 dB SDR. This suggests that the practical gains are not solely attributable to the proposed posterior-score approximation. The paper should discuss which components carry the performance and whether the likelihood score itself contributes beyond these heuristics, ideally with an ablation that keeps the heuristics but removes only the FCP-based likelihood term.
minor comments (6)
  1. [Algorithm 2, lines 9-14] The pseudocode uses IVA-initialized relative RIRs G_{1->c} while the source model requires G_{0->c} from virtual sources; the intended interpretation is only explained in prose in Appendix C.2 and should be marked explicitly in the algorithm.
  2. [Appendix C.2, Eqs. 57-58] The notation xi_1(tau) and xi_2(tau) is not accurate because the right-hand sides depend on the gradient they weight; please rename or clearly define them as step-dependent normalizers.
  3. [Eq. 37] The claim that STFT preserves the l2 norm requires a specific window and normalization; please state the normalization or cite a Parseval-type theorem for the chosen STFT setup.
  4. [References, Section 5] Several references are incomplete, including 'Lutati et al.' with no author list and a number of arXiv entries missing titles or venues; please complete them.
  5. [Tables 1 and 2] The reported standard deviations are over the five samples per mixture, not over the test set, so it is unclear whether the differences from baselines are statistically significant; consider reporting test-set confidence intervals or significance tests.
  6. [Appendix K] The hyperparameter ablations are run on only 50 validation utterances; this limitation should be stated in the main text whenever robustness is claimed.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular reduction: reported metrics are external and the FCP/ML theorem gap is a proof issue, not an input/output identity.

full rationale

ArrayDPS's central objective is to sample from p(s_1:K | x_1:C) using a pretrained single-speaker diffusion prior plus an approximated likelihood. The reported SDR/SI-SDR values are computed against ground-truth source images after IVA initialization, FCP-based filtering, and ODE sampling; none of these quantities is fitted to the evaluation targets. The 'maximum likelihood' sample-selection variant (rows 4a/4b) chooses among generated samples via mixture reconstruction SNR, not via the benchmark metrics, so it is not a fitted parameter renamed as a prediction. The diffusion prior is trained on LibriTTS, disjoint from the WSJ0 test sets, so the unsupervised claim is not definitional. FCP is cited from prior work by a co-author, but the paper attempts an independent derivation in Appendix B.2; that derivation has a load-bearing gap: the step from the joint ML objective (Eq. 44) to per-source FCP objectives (Eq. 45) is asserted via 'the orthogonal principle ... independence of the sources, and linearity of the filtering operation' without controlling realized cross-correlations, and the implemented FCP weight (Eq. 14) differs from the theorem's assumed lambda = 1/(2 sigma_N^2). This is a proof/support gap and a correctness risk, not a circular reduction: Eq. 44 and Eq. 45 are not equal by construction, and the theorem's conclusion is not the input to the pipeline. The empirical gains are externally evaluated against unseen ground-truth sources, so no step in the claimed derivation chain reduces to its own inputs.

Assumptions & free parameters 6 free parameters · 5 assumptions · 1 invented entities

The ledger lists the hand-tuned sampling weights, the signal-model and delta-posterior assumptions, and the virtual-source construct that ArrayDPS relies on. The main free parameters are sampler and guidance weights tuned on validation sets; the main axioms are the linear relative-RIR mixture model and the point-estimate approximation of the RIR posterior.

free parameters (6)
  • Mixture likelihood guidance weight xi = 2 (default), 6 (without IVA init)
    Scales the likelihood score in Eq. (57); tuned on validation and directly controls how strongly the samples must match the mixture.
  • Initial diffusion noise level tau_max = 0.8 (default), 2 (without IVA init)
    Sets the starting point of the sampling ODE; tuned on validation and shifts the balance between prior and likelihood.
  • Reference channel guidance weight lambda = 1.3
    Empirical weight for the reference-channel guidance term in Algorithm 2; theoretically unnecessary but needed for stability and performance.
  • Guidance step thresholds Nref and Nfg = Nref=200, Nfg=100 (default)
    Control how many early diffusion steps use reference guidance and IVA-initialized relative RIRs; tuned on validation.
  • FCP regularization epsilon and filter support lengths F, P = epsilon=0.001; F and P not fully explicit in text
    Epsilon smooths high-energy STFT bins in Eq. (14); F and P set the relative-RIR filter length; both affect the quality of the estimated filters.
  • Stochastic sampler parameters S_churn and S_noise = S_churn=30, S_noise=1 (default)
    EDM sampler stochasticity settings; tuned for sampling stability and final audio quality.
assumptions (5)
  • standard math Score-based diffusion/EDM sampling with Tweedie's formula yields a valid speech prior score.
    Adopted from Song et al. and Karras et al.; not proved in this paper and assumed as background.
  • domain assumption The relative-RIR virtual-source mixture model in Eqs. (9)-(11) accurately describes reverberant multi-channel speech.
    All filtering and likelihood estimation is built on this linear model; the oracle FCP ablation in Table 3 gives indirect support.
  • domain assumption Source independence and the 'orthogonal principle' let the joint multi-source filter estimation decompose into separate per-source FCP problems.
    Stated in Theorem B.1 without a complete proof; source independence does not guarantee uncorrelated regressors over short frames, so this is an unverified assumption.
  • ad hoc to paper The delta-function posterior over relative RIRs is adequate despite 20-30 dB measurement noise.
    Appendix B.2 shows the likelihood approximation is exact only in the zero-noise limit, yet the method uses it for real and simulated noisy mixtures.
  • ad hoc to paper Reference-channel guidance is a useful empirical stabilizer even though it is not part of the probabilistic model.
    Algorithm 2 lines 17-19; theoretically xi_2 is zero, but ablations show this term improves SI-SDR and reduces variance.
invented entities (1)
  • Virtual source signal s_k
    purpose: Acts as a common anchor for each speaker; all microphone channels are generated by filtering this source with estimated relative RIRs.
    An internal modeling device, not a measured or independently checkable physical quantity; final separated signals are filtered to the reference channel for evaluation and listening.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ArrayDPS: Unsupervised Blind Speech Separation with a Diffusion Prior." pith.science (2026). https://pith.science/paper/7AQRWSXT

@misc{pith2026250505657,
  author       = {Pith},
  title        = {Pith review of: ArrayDPS: Unsupervised Blind Speech Separation with a Diffusion Prior},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7AQRWSXT}},
  note         = {Machine review of arXiv:2505.05657}
}
read the original abstract

Blind Speech Separation (BSS) aims to separate multiple speech sources from audio mixtures recorded by a microphone array. The problem is challenging because it is a blind inverse problem, i.e., the microphone array geometry, the room impulse response (RIR), and the speech sources, are all unknown. We propose ArrayDPS to solve the BSS problem in an unsupervised, array-agnostic, and generative manner. The core idea builds on diffusion posterior sampling (DPS), but unlike DPS where the likelihood is tractable, ArrayDPS must approximate the likelihood by formulating a separate optimization problem. The solution to the optimization approximates room acoustics and the relative transfer functions between microphones. These approximations, along with the diffusion priors, iterate through the ArrayDPS sampling process and ultimately yield separated voice sources. We only need a simple single-speaker speech diffusion model as a prior along with the mixtures recorded at the microphones; no microphone array information is necessary. Evaluation results show that ArrayDPS outperforms all baseline unsupervised methods while being comparable to supervised methods in terms of SDR. Audio demos are provided at: https://arraydps.github.io/ArrayDPSDemo/.

Figures

Figures reproduced from arXiv: 2505.05657 by the authors.

Figure 1
Figure 1. (a) Signal mixing in the real world; (b) Relative RIR model; and (c) Relative RIR from virtual sources to real channels. Measurement noise is ignored in the figures. where hk,c(t) is the RIR from the k th source location to the c th microphone, and ∗ is the convolution operation. nc(t) is the c th microphone’s white noise [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. An overview of ArrayDPS for K = 2 sources. Left figure (a) shows the pipeline for separation with diffusion posterior sampling, which uses the likelihood approximation pipeline shown in the right figure (b). obtain sk,c2 (t) if hk,c1 (t) is invertible: g k c1→c2 (t) = hk,c2 (t) ∗ h −1 k,c1 (t) (3) sk,c2 (t) = g k c1→c2 (t) ∗ sk,c1 (t) (4) Eq. 3 is called the relative RIR between channel c1 and c2 for speaker k, and … view at source ↗
Figure 3
Figure 3. SMS-WSJ Sample 0 442c040o 443c040g Visualization [PITH_FULL_IMAGE:figures/full_fig_p028_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: SMS-WSJ Sample 1015 446c0415 442c040c Visualization 28 [PITH_FULL_IMAGE:figures/full_fig_p028_4.png]
Figure 5
Figure 5. Figure 5: SMS-WSJ Sample 1120 445c040c 441c040m Visualization [PITH_FULL_IMAGE:figures/full_fig_p029_5.png]
Figure 6
Figure 6. Figure 6: SMS-WSJ Sample 999 441c040c 447c040k Visualization 29 [PITH_FULL_IMAGE:figures/full_fig_p029_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

82 extracted references · 53 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    Allen, J. B. and Berkley, D. A. Image method for efficiently simulating small-room acoustics. The Journal of the Acoustical Society of America, 65 0 (4): 0 943--950, 1979

    1979

  3. [3]

    A new learning algorithm for blind signal separation

    Amari, S.-i., Cichocki, A., and Yang, H. A new learning algorithm for blind signal separation. Advances in neural information processing systems, 8, 1995

    work page 1995

  4. [4]

    Reverberation as supervision for speech separation

    Aralikatti, R., Boeddeker, C., Wichern, G., Subramanian, A., and Le Roux, J. Reverberation as supervision for speech separation. In IEEE International Conference on Acoustics, Speech and Signal Processing, pp.\ 1--5, 2023

    work page 2023

  5. [5]

    Blind inversion using latent diffusion priors, 2024

    Bai, W., Chen, S., Chen, W., and Sun, H. Blind inversion using latent diffusion priors, 2024. URL https://arxiv.org/abs/2407.01027

    arXiv 2024

  6. [6]

    A., Fontaine, M., and Yoshii, K

    Bando, Y., Sekiguchi, K., Masuyama, Y., Nugraha, A. A., Fontaine, M., and Yoshii, K. Neural full-rank spatial covariance analysis for blind source separation. IEEE Signal Processing Letters, 28: 0 1670--1674, 2021

    work page 2021

  7. [7]

    Weakly-supervised neural full-rank spatial covariance analysis for a front-end system of distant speech recognition

    Bando, Y., Aizawa, T., Itoyama, K., and Nakadai, K. Weakly-supervised neural full-rank spatial covariance analysis for a front-end system of distant speech recognition. In Interspeech, pp.\ 3824--3828, 2022

    work page 2022

  8. [8]

    Example of mixture model

    Boeddeker, C. Example of mixture model. https://github.com/fgnt/pb_bss/blob/master/examples/mixture_model_example.ipynb, 2019

    work page 2019

Show all 82 references
  1. [9]

    Cherry, E. C. Some experiments on the recognition of speech, with one and with two ears. The Journal of the acoustical society of America, 25 0 (5): 0 975--979, 1953

    1953

  2. [10]

    Chung, H., Kim, J., Kim, S., and Ye, J. C. Parallel diffusion models of operator and image for blind inverse problems. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 6059--6069, 2023 a

    2023

  3. [11]

    T., Klasky, M

    Chung, H., Kim, J., Mccann, M. T., Klasky, M. L., and Ye, J. C. Diffusion posterior sampling for general noisy inverse problems. In International Conference on Learning Representations, 2023 b . URL https://openreview.net/forum?id=OnD9zGAGT0k

    2023

  4. [12]

    Unsupervised training of a deep clustering model for multichannel blind source separation

    Drude, L., Hasenklever, D., and Haeb-Umbach, R. Unsupervised training of a deep clustering model for multichannel blind source separation. In ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 695--699. IEEE, 2019 a

    2019

  5. [13]

    Sms-wsj: Database, performance measures, and baseline recipe for multi-channel source separation and recognition, 2019 b

    Drude, L., Heitkaemper, J., Boeddeker, C., and Haeb-Umbach, R. Sms-wsj: Database, performance measures, and baseline recipe for multi-channel source separation and recognition, 2019 b . URL https://arxiv.org/abs/1910.13934

    2019 arXiv

  6. [14]

    Unsupervised training of neural mask-based beamforming

    Drude, L., Heymann, J., and Haeb-Umbach, R. Unsupervised training of neural mask-based beamforming. arXiv preprint arXiv:1904.01578, 2019 c

    1904 arXiv

  7. [15]

    A consolidated perspective on multimicrophone speech enhancement and source separation

    Gannot, S., Vincent, E., Markovich-Golan, S., and Ozerov, A. A consolidated perspective on multimicrophone speech enhancement and source separation. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 25 0 (4): 0 692--730, 2017. doi:10.1109/TASLP.2016.2647702

    2017

  8. [16]

    Han, C., Wilson, K., Wisdom, S., and Hershey, J. R. Unsupervised multi-channel separation and adaptation. In ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 721--725. IEEE, 2024

    2024

  9. [17]

    R., Chen, Z., Le Roux, J., and Watanabe, S

    Hershey, J. R., Chen, Z., Le Roux, J., and Watanabe, S. Deep clustering: Discriminative embeddings for segmentation and separation. In 2016 IEEE international conference on acoustics, speech and signal processing (ICASSP), pp.\ 31--35. IEEE, 2016

    2016

  10. [18]

    Solution of permutation problem in frequency domain ica, using multivariate probability density functions

    Hiroe, A. Solution of permutation problem in frequency domain ica, using multivariate probability density functions. In Rosca, J., Erdogmus, D., Pr \'i ncipe, J. C., and Haykin, S. (eds.), Independent Component Analysis and Blind Signal Separation, pp.\ 601--608, Berlin, Heide...

    2006

  11. [19]

    Denoising diffusion probabilistic models

    Ho, J., Jain, A., and Abbeel, P. Denoising diffusion probabilistic models. arXiv preprint arxiv:2006.11239, 2020

    2006 arXiv

  12. [20]

    a rinen, A., Hurri, J., Hoyer, P. O., Hyv \

    Hyv \"a rinen, A., Hurri, J., Hoyer, P. O., Hyv \"a rinen, A., Hurri, J., and Hoyer, P. O. Independent component analysis. Springer, 2009

    2009

  13. [21]

    Complex angular central gaussian mixture model for directional statistics in mask-based microphone array signal processing

    Ito, N., Araki, S., and Nakatani, T. Complex angular central gaussian mixture model for directional statistics in mask-based microphone array signal processing. In 2016 24th European Signal Processing Conference (EUSIPCO), pp.\ 1153--1157, 2016. doi:10.1109/EUSIPCO.2016.7760429

    2016

  14. [22]

    and Thickstun, J

    Jayaram, V. and Thickstun, J. Source separation with deep generative priors. In International Conference on Machine Learning, pp.\ 4724--4735. PMLR, 2020

    2020

  15. [23]

    Kalkhorani, V. A. and Wang, D. Tf-crossnet: Leveraging global, cross-band, narrow-band, and positional encoding for single- and multi-channel speaker separation. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 32: 0 4999--5009, 2024. doi:10.1109/TASLP.2024.3492803

    2024

  16. [24]

    R., and Zhang, S

    Karchkhadze, T., Izadi, M. R., and Zhang, S. Improving source extraction with diffusion and consistency models. arXiv preprint arXiv:2412.06965, 2024

    2024 arXiv

  17. [25]

    Elucidating the design space of diffusion-based generative models

    Karras, T., Aittala, M., Aila, T., and Laine, S. Elucidating the design space of diffusion-based generative models. In Proc. NeurIPS, 2022

    2022

  18. [26]

    Independent vector analysis: An extension of ica to multivariate components

    Kim, T., Eltoft, T., and Lee, T.-W. Independent vector analysis: An extension of ica to multivariate components. In International conference on independent component analysis and signal separation, pp.\ 165--172. Springer, 2006

    2006

  19. [27]

    The reverb challenge: A common evaluation framework for dereverberation and recognition of reverberant speech

    Kinoshita, K., Delcroix, M., Yoshioka, T., Nakatani, T., Habets, E., Haeb-Umbach, R., Leutnant, V., Sehr, A., Kellermann, W., Maas, R., Gannot, S., and Raj, B. The reverb challenge: A common evaluation framework for dereverberation and recognition of reverberant speech. In 201...

    2013

  20. [28]

    L., and Khudanpur, S

    Ko, T., Peddinti, V., Povey, D., Seltzer, M. L., and Khudanpur, S. A study on data augmentation of reverberant speech for robust speech recognition. In 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 5220--5224, 2017. doi:10.1109/IC...

    2017

  21. [29]

    Fast diffusion em: a diffusion model for blind inverse problems with application to deconvolution

    Laroche, C., Almansa, A., and Coupete, E. Fast diffusion em: a diffusion model for blind inverse problems with application to deconvolution. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp.\ 5271--5281, 2024

    2024

  22. [30]

    Tabe: Decoupling spatial and spectral processing with taylor’s unfolding method in the beamspace domain for multi-channel speech enhancement

    Li, A., Yu, G., Xu, Z., Fan, C., Li, X., and Zheng, C. Tabe: Decoupling spatial and spectral processing with taylor’s unfolding method in the beamspace domain for multi-channel speech enhancement. Information Fusion, 101: 0 101976, 2024

    2024

  23. [31]

    End-to-end microphone permutation and number invariant multi-channel speech separation

    Luo, Y., Chen, Z., Mesgarani, N., and Yoshioka, T. End-to-end microphone permutation and number invariant multi-channel speech separation. In ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 6394--6398. IEEE, 2020

    2020

  24. [32]

    Separate and diffuse: Using a pretrained diffusion model for better source separation

    Lutati, S., Nachmani, E., and Wolf, L. Separate and diffuse: Using a pretrained diffusion model for better source separation. In The Twelfth International Conference on Learning Representations

  25. [33]

    Multi-source diffusion models for simultaneous music generation and separation

    Mariani, G., Tallini, I., Postolache, E., Mancusi, M., Cosmo, L., and Rodol \`a , E. Multi-source diffusion models for simultaneous music generation and separation. In The Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=h...

    2024

  26. [34]

    McDermott, J. H. The cocktail party problem. Current Biology, 19 0 (22): 0 R1024--R1027, 2009

    2009

  27. [35]

    Solving audio inverse problems with a diffusion model

    Moliner, E., Lehtinen, J., and Välimäki, V. Solving audio inverse problems with a diffusion model. In ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 1--5, 2023. doi:10.1109/ICASSP49357.2023.10095637

    2023

  28. [36]

    Blind audio bandwidth extension: A diffusion-based zero-shot approach

    Moliner, E., Elvander, F., and Välimäki, V. Blind audio bandwidth extension: A diffusion-based zero-shot approach. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 32: 0 5092--5105, 2024 a . doi:10.1109/TASLP.2024.3507566

    2024

  29. [37]

    Buddy: Single-channel blind unsupervised dereverberation with diffusion models

    Moliner, E., Lemercier, J.-M., Welker, S., Gerkmann, T., and Välimäki, V. Buddy: Single-channel blind unsupervised dereverberation with diffusion models. In 2024 18th International Workshop on Acoustic Signal Enhancement (IWAENC), pp.\ 120--124, 2024 b . doi:10.1109/IWAENC6148...

    2024

  30. [38]

    Unsupervised estimation of nonlinear audio effects: Comparing diffusion-based and adversarial approaches, 2025

    Moliner, E., Švento, M., Wright, A., Juvela, L., Rajmic, P., and Välimäki, V. Unsupervised estimation of nonlinear audio effects: Comparing diffusion-based and adversarial approaches, 2025. URL https://arxiv.org/abs/2504.04751

    2025

  31. [39]

    Joint dereverberation and separation with iterative source steering

    Nakashima, T., Scheibler, R., Togami, M., and Ono, N. Joint dereverberation and separation with iterative source steering. ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 216--220, 2021. URL https://api.semanticscholar...

    2021

  32. [40]

    and Wang, D

    Pandey, A. and Wang, D. On cross-corpus generalization of deep learning based speech enhancement. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 28: 0 2489--2499, 2020. URL https://api.semanticscholar.org/CorpusID:211068769

    2020

  33. [41]

    and Li, X

    Quan, C. and Li, X. Spatialnet: Extensively learning spatial information for multichannel joint speech separation, denoising and dereverberation, 2023. URL https://arxiv.org/abs/2307.16516

    2023 arXiv

  34. [42]

    The DUET Blind Source Separation Algorithm, pp.\ 217--241

    Rickard, S. The DUET Blind Source Separation Algorithm, pp.\ 217--241. Springer Netherlands, Dordrecht, 2007. ISBN 978-1-4020-6479-1. doi:10.1007/978-1-4020-6479-1_8. URL https://doi.org/10.1007/978-1-4020-6479-1_8

    2007 doi

  35. [43]

    Perceptual evaluation of speech quality (pesq)-a new method for speech quality assessment of telephone networks and codecs

    Rix, A., Beerends, J., Hollier, M., and Hekstra, A. Perceptual evaluation of speech quality (pesq)-a new method for speech quality assessment of telephone networks and codecs. In 2001 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings (Cat. ...

    2001

  36. [44]

    L., Wisdom, S., Erdogan, H., and Hershey, J

    Roux, J. L., Wisdom, S., Erdogan, H., and Hershey, J. R. Sdr – half-baked or well done? ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 626--630, 2018. URL https://api.semanticscholar.org/CorpusID:53246666

    2019

  37. [45]

    and Ogawa, T

    Saijo, K. and Ogawa, T. Self-remixing: Unsupervised speech separation via separation and remixing. In ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 1--5. IEEE, 2023

    2023

  38. [46]

    G., Pan, Z., and Le Roux , J

    Saijo, K., Wichern, G., Germain, F. G., Pan, Z., and Le Roux , J. Enhanced reverberation as supervision for unsupervised speech separation. In Interspeech 2024, pp.\ 607--611, 2024. doi:10.21437/Interspeech.2024-1241

    2024 doi

  39. [47]

    Underdetermined convolutive blind source separation via frequency bin-wise clustering and permutation alignment

    Sawada, H., Araki, S., and Makino, S. Underdetermined convolutive blind source separation via frequency bin-wise clustering and permutation alignment. IEEE Transactions on Audio, Speech, and Language Processing, 19 0 (3): 0 516--527, 2011. doi:10.1109/TASL.2010.2051355

    2011

  40. [48]

    A review of blind source separation methods: two converging routes to ilrma originating from ica and nmf

    Sawada, H., Ono, N., Kameoka, H., Kitamura, D., and Saruwatari, H. A review of blind source separation methods: two converging routes to ilrma originating from ica and nmf. APSIPA Transactions on Signal and Information Processing, 8: 0 e12, 2019

    2019

  41. [49]

    and Saijo, K

    Scheibler, R. and Saijo, K. torchiva: Tools for independent vector analysis in pytorch. https://github.com/fakufaku/torchiva, 2022

    2022

  42. [50]

    Diffusion-based generative speech source separation

    Scheibler, R., Ji, Y., Chung, S.-W., Byun, J., Choe, S., and Choi, M.-S. Diffusion-based generative speech source separation. In ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 1--5. IEEE, 2023

    2023

  43. [51]

    Bootstrapping single-channel source separation via unsupervised spatial clustering on stereo mixtures

    Seetharaman, P., Wichern, G., Le Roux, J., and Pardo, B. Bootstrapping single-channel source separation via unsupervised spatial clustering on stereo mixtures. In ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 356--360....

    2019

  44. [52]

    Sivaraman, A., Wisdom, S., Erdogan, H., and Hershey, J. R. Adapting speech separation to real-world meetings using mixture invariant training. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 686--690. IEEE, 2022

    2022

  45. [53]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Sohl-Dickstein, J., Weiss, E., Maheswaranathan, N., and Ganguli, S. Deep unsupervised learning using nonequilibrium thermodynamics. In International conference on machine learning, pp.\ 2256--2265. PMLR, 2015

    2015

  46. [54]

    Pseudoinverse-guided diffusion models for inverse problems

    Song, J., Vahdat, A., Mardani, M., and Kautz, J. Pseudoinverse-guided diffusion models for inverse problems. In International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=9_gsMA8MRKQ

    2023

  47. [55]

    P., Kumar, A., Ermon, S., and Poole, B

    Song, Y., Sohl-Dickstein, J., Kingma, D. P., Kumar, A., Ermon, S., and Poole, B. Score-based generative modeling through stochastic differential equations. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id=PxTIG12RRHS

    2021

  48. [56]

    Subakan, Y. C. and Smaragdis, P. Generative adversarial source separation. In 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 26--30. IEEE, 2018

    2018

  49. [58]

    Estimation and restoration of unknown nonlinear distortion using diffusion

    S vento, M., Moliner, E., Juvela, L., Wright, A., and V \"a lim \"a ki, V. Estimation and restoration of unknown nonlinear distortion using diffusion. arXiv preprint arXiv:2501.05959, 2025 b

    2025 arXiv

  50. [59]

    H., Hendriks, R

    Taal, C. H., Hendriks, R. C., Heusdens, R., and Jensen, J. An algorithm for intelligibility prediction of time–frequency weighted noisy speech. IEEE Transactions on Audio, Speech, and Language Processing, 19 0 (7): 0 2125--2136, 2011. doi:10.1109/TASL.2011.2114881

    2011

  51. [60]

    E., Yoshioka, T., Wang, H., Chen, Z., and Huang, X

    Taherian, H., Eskimez, S. E., Yoshioka, T., Wang, H., Chen, Z., and Huang, X. One model to enhance them all: array geometry agnostic multi-channel personalized speech enhancement. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICA...

    2022

  52. [61]

    Multi-input multi-output complex spectral mapping for speaker separation

    Taherian, H., Pandey, A., Wong, D., Xu, B., and Wang, D. Multi-input multi-output complex spectral mapping for speaker separation. ISCA, 2023

    2023

  53. [62]

    Leveraging sound localization to improve continuous speaker separation

    Taherian, H., Pandey, A., Wong, D., Xu, B., and Wang, D. Leveraging sound localization to improve continuous speaker separation. In ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 621--625, 2024. doi:10.1109/ICASSP4848...

    2024

  54. [63]

    Hrtf estimation using a score-based prior

    Thuillier, E., Lemercier, J.-M., Moliner, E., Gerkmann, T., and V \"a lim \"a ki, V. Hrtf estimation using a score-based prior. In ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 1--5. IEEE, 2025

    2025

  55. [64]

    Unsupervised training for deep speech source separation with kullback-leibler divergence based probabilistic loss function

    Togami, M., Masuyama, Y., Komatsu, T., and Nakagome, Y. Unsupervised training for deep speech source separation with kullback-leibler divergence based probabilistic loss function. In ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICA...

    2020

  56. [65]

    Tran Vu, D. H. and Haeb-Umbach, R. Blind speech separation employing directional statistics in an expectation maximization framework. In 2010 IEEE International Conference on Acoustics, Speech and Signal Processing, pp.\ 241--244, 2010. doi:10.1109/ICASSP.2010.5495994

    2010

  57. [66]

    Unsupervised deep clustering for source separation: Direct learning from mixtures using spatial information

    Tzinis, E., Venkataramani, S., and Smaragdis, P. Unsupervised deep clustering for source separation: Direct learning from mixtures using spatial information. In ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 81--85. IEEE, 2019

    2019

  58. [67]

    P., and Hershey, J

    Tzinis, E., Wisdom, S., Jansen, A., Hershey, S., Remez, T., Ellis, D. P., and Hershey, J. R. Into the wild with audioscope: Unsupervised audio-visual separation of on-screen sounds. arXiv preprint arXiv:2011.01143, 2020

    2011 arXiv

  59. [68]

    K., Xu, B., Smaragdis, P., and Kumar, A

    Tzinis, E., Adi, Y., Ithapu, V. K., Xu, B., Smaragdis, P., and Kumar, A. Remixit: Continual self-training of speech enhancement models via bootstrapped remixing. IEEE Journal of Selected Topics in Signal Processing, 16 0 (6): 0 1329--1341, 2022 a

    2022

  60. [69]

    Tzinis, E., Wisdom, S., Remez, T., and Hershey, J. R. Audioscopev2: Audio-visual attention architectures for calibrated open-domain on-screen sound separation. In European Conference on Computer Vision, pp.\ 368--385. Springer, 2022 b

    2022

  61. [70]

    Performance measurement in blind audio source separation

    Vincent, E., Gribonval, R., and Fevotte, C. Performance measurement in blind audio source separation. IEEE Transactions on Audio, Speech, and Language Processing, 14 0 (4): 0 1462--1469, 2006. doi:10.1109/TSA.2005.858005

    2006

  62. [71]

    and Chen, J

    Wang, D. and Chen, J. Supervised speech separation based on deep learning: An overview. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 26 0 (10): 0 1702--1726, 2018. doi:10.1109/TASLP.2018.2842159

    2018

  63. [72]

    and Watanabe, S

    Wang, Z.-Q. and Watanabe, S. UNSSOR : Unsupervised neural speech separation by leveraging over-determined training mixtures. In Thirty-seventh Conference on Neural Information Processing Systems, 2023. URL https://openreview.net/forum?id=T5h69frFF7

    2023

  64. [73]

    Wang, Z.-Q., Le Roux, J., and Hershey, J. R. Multi-channel deep clustering: Discriminative spectral and spatial embeddings for speaker-independent speech separation. In 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 1--5, 2018. doi...

    2018

  65. [75]

    Wang, Z.-Q., Wichern, G., and Roux, J. L. Convolutive prediction for monaural speech dereverberation and noisy-reverberant speaker separation. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 29: 0 3476--3490, 2021 b . doi:10.1109/TASLP.2021.3129363

    2021

  66. [76]

    Tf-gridnet: Integrating full- and sub-band modeling for speech separation

    Wang, Z.-Q., Cornell, S., Choi, S., Lee, Y., Kim, B.-Y., and Watanabe, S. Tf-gridnet: Integrating full- and sub-band modeling for speech separation. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 31: 0 3221--3236, 2023. doi:10.1109/TASLP.2023.3304482

    2023

  67. [77]

    Unsupervised sound separation using mixture invariant training

    Wisdom, S., Tzinis, E., Erdogan, H., Weiss, R., Wilson, K., and Hershey, J. Unsupervised sound separation using mixture invariant training. Advances in neural information processing systems, 33: 0 3846--3857, 2020

    2020

  68. [78]

    and Choudhury, R

    Xu, A. and Choudhury, R. R. Learning to separate voices by spatial regions. In International Conference on Machine Learning, pp.\ 24539--24549. PMLR, 2022

    2022

  69. [79]

    Fovnet: Configurable field-of-view speech enhancement with low computation and distortion for smart glasses

    Xu, Z., Aroudi, A., Tan, K., Pandey, A., Lee, J.-S., Xu, B., and Nesta, F. Fovnet: Configurable field-of-view speech enhancement with low computation and distortion for smart glasses. In Interspeech 2024, pp.\ 3350--3354, 2024. doi:10.21437/Interspeech.2024-2124

    2024 doi

  70. [80]

    Vararray: Array-geometry-agnostic continuous speech separation

    Yoshioka, T., Wang, X., Wang, D., Tang, M., Zhu, Z., Chen, Z., and Kanda, N. Vararray: Array-geometry-agnostic continuous speech separation. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 6027--6031. IEEE, 2022

    2022

  71. [81]

    Zero-shot duet singing voices separation with diffusion models

    Yu, C.-Y., Postolache, E., Rodol \`a , E., and Fazekas, G. Zero-shot duet singing voices separation with diffusion models. arXiv preprint arXiv:2311.07345, 2023

    2023 arXiv

  72. [82]

    Permutation invariant training of deep models for speaker-independent multi-talker speech separation

    Yu, D., Kolbæk, M., Tan, Z.-H., and Jensen, J. Permutation invariant training of deep models for speaker-independent multi-talker speech separation. In 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 241--245, 2017. doi:10.1109/ICAS...

    2017

  73. [83]

    J., Jia, Y., Chen, Z., and Wu, Y

    Zen, H., Dang, V., Clark, R., Zhang, Y., Weiss, R. J., Jia, Y., Chen, Z., and Wu, Y. Libritts: A corpus derived from librispeech for text-to-speech. In Interspeech 2019, pp.\ 1526--1530, 2019. doi:10.21437/Interspeech.2019-2441

    2019 doi

  74. [84]

    Closing the gap between time-domain multi-channel speech enhancement on real and simulation conditions

    Zhang, W., Shi, J., Li, C., Watanabe, S., and Qian, Y. Closing the gap between time-domain multi-channel speech enhancement on real and simulation conditions. In 2021 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA), pp.\ 146--150, 2021. doi:1...

    2021

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.