Pith. sign in

REVIEW 4 major objections 5 minor 2 cited by

ClearerVoice-Studio: Bridging Advanced Speech Processing Research and Practical Deployment

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read ClearerVoice-Studio is an open-source speech processing toolkit that unifies enhancement, separation, super-resolution, and multimodal target-speaker extraction, and its new face-conditioned model outperforms the state of the art on two…

desk verdict A useful, honest toolkit paper whose headline 'state-of-the-art' benchmarks are not yet controlled; worth reviewing with revision. read the letter →

arxiv 2506.19398 v1 pith:2NTYGFV4 submitted 2025-06-24 cs.SD eess.AS

Shengkui Zhao , Zexu Pan , Bin Ma This is my paper
classification cs.SDeess.AS
keywords speechenhancementseparationsuper-resolutionmultimodaltargetspeakerextractionaudio-visualopen-sourcetoolkitMossFormer2VoxCeleb2benchmark
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

This paper presents ClearerVoice-Studio, an open-source toolkit that brings together four interconnected speech tasks — enhancement, separation, super-resolution, and multimodal target-speaker extraction — under one platform with pretrained models, training scripts, and evaluation tools. The paper argues that this convergence matters because real-world recordings often suffer from several of these degradations at once, and existing general-purpose or single-task toolkits do not cover the combined scenario. Its strongest technical claim is a new face-conditioned extraction model, AV-MossFormer2 TSE 16K, which it reports outperforms the state-of-the-art AV-TFGridNet on both two-speaker and three-speaker mixtures of the new VoxCeleb2 benchmark it introduces. If correct, the toolkit gives researchers and practitioners a single, practical starting point for robust speech processing across these tasks.

What carries the argument

The load-bearing mechanism is MossFormer2, a hybrid network combining a gated single-head Transformer with convolution-augmented joint self-attentions and a recurrent module, which serves as the feature-mapping backbone for enhancement, separation, super-resolution, and the new extraction model. For the new AV-MossFormer2 TSE 16K, a visual encoder's frame-level features are repeatedly concatenated and projected into the MossFormer2 recurrent module, fusing lip and face information with speech features; this fusion is what lets the model isolate a target speaker from mixtures.

What would settle it

Run AV-TFGridNet and the other Table 5 baselines on the exact same VoxCeleb2 2-mix and 3-mix test sets with the same evaluation script, then check whether AV-MossFormer2 still reports higher SI-SNRi and SNRi.

Watch

Extended reading notes

Core claim

The central claim is that ClearerVoice-Studio provides state-of-the-art pretrained models for all four tasks, built on two core architectures — FRCRN for a 16 kHz enhancement model and MossFormer2 as the shared backbone for the remaining models — and that the newly proposed AV-MossFormer2 TSE 16K, which extends the MossFormer2 separation network with repeated frame-level visual feature fusion, achieves higher SI-SNRi and SNRi than the previous state-of-the-art face-conditioned extractor on both 2-speaker and 3-speaker mixtures. The paper also introduces a standardized VoxCeleb2 mixture benchmark for audio-visual target speaker extraction and reports community adoption metrics as evidence of practical value.

Load-bearing premise

The paper's competitive-performance claim rests on comparing its own evaluations against published baseline numbers that were produced under different datasets, sampling rates, and protocols, without re-running the baselines under identical conditions.

Editorial extensions

If this is right

  • A single toolkit can compose tasks, e.g., enhancement before super-resolution, to handle realistic multi-degradation audio.
  • The reported AV-MossFormer2 results, if reproduced, establish a new state of the art for face-conditioned extraction on the VoxCeleb2 benchmark.
  • Researchers get reproducible training and fine-tuning scripts and a unified evaluation resource instead of stitching together task-specific repositories.
  • Practitioners can run pretrained models on common audio formats at multiple sampling rates without task-specific engineering.

Reading between the lines

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

  • The VoxCeleb2 mixture benchmark, if adopted by the community, could become a standard test bed for face-conditioned extraction, making future comparisons direct.
  • Because the same MossFormer2 backbone carries four tasks, improvements in the backbone could transfer across enhancement, separation, super-resolution, and extraction simultaneously.
  • The toolkit's design suggests a path where composite pipelines (for example, enhance-then-separate-then-upsample) become the default real-world workflow, a direction the paper mentions but does not evaluate empirically.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. This manuscript presents ClearerVoice-Studio, an open-source speech-processing toolkit covering speech enhancement, speech separation, speech super-resolution, and audio-visual/multimodal target speaker extraction. The toolkit builds on the authors' earlier FRCRN and MossFormer2 architectures and adds a new audio-visual speech extraction model, AV-MossFormer2 TSE 16K. The paper describes the system architecture, training data and losses, reports benchmark results for each task, and summarizes community adoption metrics such as GitHub stars and forks.

Significance. If the reported benchmark results are reliable, the toolkit is a practically useful contribution: it provides pretrained models, training scripts, the SpeechScore evaluation suite, and interactive interfaces, and it is open source with meaningful community uptake. The most scientifically interesting component is the new AV-MossFormer2 model and the introduced VoxCeleb2 mixture benchmark. The paper's central competitive claim, however, rests on comparisons against published numbers obtained under different protocols, so the significance of the SOTA claim is currently conditional on controlled re-evaluation.

major comments (4)
  1. [§3.2, Table 5]
  2. [§3.2, Table 3]
  3. [§3.2, Table 4]
  4. [§3.1.1 and §3.2 (AV-MossFormer2)]
minor comments (5)
  1. [Abstract vs. §4]
  2. [§3.1.1]
  3. [Table 3]
  4. [§3.2, Table 3]
  5. [§2.2.3]

Circularity Check

0 steps flagged · score 2.0 of 10

ClearerVoice-Studio is a toolkit paper with no definitional or equation-level circularity; its flagship AV-MossFormer2-versus-AV-TFGridNet comparison rests on a co-authored baseline under a self-introduced protocol, which is a comparability concern rather than a circular reduction.

full rationale

ClearerVoice-Studio is a systems and toolkit paper; it contains no derivation of a claimed result from its own inputs by construction. The core architectures (FRCRN [11], MossFormer2 [12], HiFi-SR [14]) are the authors' prior work, but each was published at peer-reviewed venues (ICASSP, ASRU), so under the standard rule these citations are independent evidence for the components rather than circular support. Single-modality evaluations in Tables 1-3 are measured against external baselines (DCCRN+, MFNet, TridentSE, SepFormer, TF-GridNet, SPMamba) on widely used benchmarks (DNS-2020, VoiceBank+DEMAND, WSJ0-2Mix, WHAM!, LRS2-2Mix), giving the toolkit claim independent content. The one load-bearing competitive statement, 'AV-MossFormer2 outperforms the state-of-the-art AV-TFGridNet' (Sec. 3.2, Table 5), is assembled from group-internal pieces: the new model is an audio-visual extension of the authors' MossFormer2, the AV-TFGridNet row is a baseline from the co-authored ASRU 2023 paper [16], and the evaluation corpus is a VoxCeleb2 mixture benchmark the paper says it 'introduced'. The paper does not state whether the AV-TFGridNet checkpoint was re-run under the identical mixture-generation, SNR range, noise, resampling, and metric protocol, so the comparison may be uncontrolled. This is a protocol-comparability and validation weakness, not circularity: the reported AV-MossFormer2 numbers are genuine measurements, no parameter is fitted to force the comparison, and the superiority claim does not reduce to an input by definition or by a self-citation chain that replaces evidence. Accordingly, no step meets the evidentiary bar for a circularity finding. Score 2 reflects the heavy self-citation surrounding the headline comparison while acknowledging the independent external evaluations elsewhere in the paper.

Assumptions & free parameters 0 free parameters · 3 assumptions · 2 invented entities

No numerical free parameters are fitted in this paper; the models' weights are learned from data. The axioms concern the comparability of external baselines, the validity of the self-created VoxCeleb2 benchmark, and the architectural fusion choice in AV-MossFormer2.

assumptions (3)
  • domain assumption Published baseline numbers from other papers are directly comparable to the authors' evaluation runs.
    Tables 3 and 5 compare against 'results from published works' without re-running baselines under identical conditions.
  • domain assumption The VoxCeleb2 mixture benchmark constructed by the authors is a valid test bed for audio-visual speaker extraction.
    Introduced in Table 5; no external validation yet.
  • ad hoc to paper AV-MossFormer2's frame-level visual feature concatenation into the MossFormer-Recurrent module is an effective fusion mechanism.
    Stated in §2.2.3 without ablations or formal justification.
invented entities (2)
  • AV-MossFormer2 TSE 16K model independent evidence
    purpose: Audio-visual target speaker extraction conditioned on face video.
    Public code and pretrained weights allow external replication; the reported superiority over AV-TFGridNet is self-evaluated.
  • VoxCeleb2 mixture benchmark
    purpose: Standardized evaluation protocol for audio-visual speaker extraction with 2-mix and 3-mix mixtures.
    Created for this paper; not yet established in the community.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ClearerVoice-Studio: Bridging Advanced Speech Processing Research and Practical Deployment." pith.science (2026). https://pith.science/paper/2NTYGFV4

@misc{pith2026250619398,
  author       = {Pith},
  title        = {Pith review of: ClearerVoice-Studio: Bridging Advanced Speech Processing Research and Practical Deployment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2NTYGFV4}},
  note         = {Machine review of arXiv:2506.19398}
}
read the original abstract

This paper introduces ClearerVoice-Studio, an open-source, AI-powered speech processing toolkit designed to bridge cutting-edge research and practical application. Unlike broad platforms like SpeechBrain and ESPnet, ClearerVoice-Studio focuses on interconnected speech tasks of speech enhancement, separation, super-resolution, and multimodal target speaker extraction. A key advantage is its state-of-the-art pretrained models, including FRCRN with 3 million uses and MossFormer with 2.5 million uses, optimized for real-world scenarios. It also offers model optimization tools, multi-format audio support, the SpeechScore evaluation toolkit, and user-friendly interfaces, catering to researchers, developers, and end-users. Its rapid adoption attracting 3000 GitHub stars and 239 forks highlights its academic and industrial impact. This paper details ClearerVoice-Studio's capabilities, architectures, training strategies, benchmarks, community impact, and future plan. Source code is available at https://github.com/modelscope/ClearerVoice-Studio.

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Where Speech Enhancement Hurts Recognition: An Inference Time Polar Projection Diagnosis

    eess.AS 2026-07 conditional novelty 6.0 of 10

    Magnitude strength, not estimated phase, drives SE-induced ASR degradation, and the optimal strength is recognizer-dependent (strong for wav2vec 2.0, mild for Whisper).

  2. CueNet: Robust Audio-Visual Speaker Extraction through Cross-Modal Cue Mining and Interaction

    cs.MM 2026-03 conditional novelty 6.0 of 10

    CueNet disentangles speaker, acoustic-synchronisation, and semantic-synchronisation cues from audio-visual speech and fuses them by estimated reliability, preserving extraction performance under visual degradation wit...

Reference graph

Works this paper leans on

45 extracted references · 40 canonical work pages · cited by 2 Pith papers

  1. [1]

    While crucial for these applications, ac- curately processing speech is challenged by the often degraded quality of real-world audio

    Introduction Speech processing is now fundamental to modern technology, powering advancements in communication, voice interfaces, and multimedia [1]. While crucial for these applications, ac- curately processing speech is challenged by the often degraded quality of real-world audio. Noise, interfering speech, re- verberation, and low resolution commonly c...

    work page

  2. [2]

    2https://github.com/resemble-ai/resemble-enhance arXiv:2506.19398v1 [cs.SD] 24 Jun 2025 2.1

    Functionalities and architectures This section details ClearerV oice-Studio’s functionalities across its target tasks and provides an in-depth look at the core model architectures that power these functionalities. 2https://github.com/resemble-ai/resemble-enhance arXiv:2506.19398v1 [cs.SD] 24 Jun 2025 2.1. ClearerV oice-Studio functionalities 2.1.1. Easy a...

    arXiv 2025

  3. [3]

    Training strategies 3.1.1

    Training strategies and performance 3.1. Training strategies 3.1.1. Dataset preparation ClearerV oice-Studio’s pre-trained models are trained on a large and diverse dataset combining public and internal resources. For speech enhancement, the fullband clean speech sources in- clude the 4th DNS-Challenge speech dataset [20] and the in- ternal TTS dataset. T...

    work page 2020

  4. [4]

    Beyond the presented evaluations, ClearerV oice- Studio is available for live demos on HuggingFace and Mod- elScope, enabling users to experiment with real-world record- ings

    Discussion and future plan We presented ClearerV oice-Studio serving as a valuable bridge between advanced speech processing research and practical ap- plications. Beyond the presented evaluations, ClearerV oice- Studio is available for live demos on HuggingFace and Mod- elScope, enabling users to experiment with real-world record- ings. We have also gene...

    work page

  5. [5]

    Deep learning for audio signal processing,

    H. Purwins, B. Li, T. Virtanen, J. Schl ¨uter, S.-Y . Chang, and T. Sainath, “Deep learning for audio signal processing,”IEEE J. Sel. Top. Signal Process., vol. 13, no. 2, pp. 206–219, 2019

    work page 2019

  6. [6]

    Mamba in Speech: Towards an alternative to self-attention,

    X. Zhang, Q. Zhang, H. Liu, T. Xiao, X. Qian, B. Ahmed, E. Am- bikairajah, H. Li, and J. Epps, “Mamba in Speech: Towards an alternative to self-attention,”preprint arXiv:2405.12609, 2025

    arXiv 2025

  7. [7]

    DeepMMSE: A deep learning approach to mmse-based noise power spectral density estimation,

    Q. Zhang, A. Nicolson, M. Wang, K. K. Paliwal, and C. Wang, “DeepMMSE: A deep learning approach to mmse-based noise power spectral density estimation,”IEEE/ACM Trans. Audio Speech Lang. Process., vol. 28, pp. 1404–1415, 2020

    work page 2020

  8. [8]

    SpeechBrain: A general-purpose speech toolkit,

    M. Ravanelli, T. Parcollet, P. Plantinga, A. Rouhe, S. Cornell, L. Lugosch, C. Subakan, N. Dawalatabad, A. Heba, J. Zhong, J.-C. Chou, S.-L. Yeh, S.-W. Fu, C.-F. Liao, E. Rastorgueva, F. Grondin, W. Aris, H. Na, Y . Gao, R. D. Mori, and Y . Bengio, “SpeechBrain: A general-purpose speech toolkit,”

    work page

Show all 45 references
  1. [9]

    AudioSR: Versatile audio super-resolution at scale,

    H. Liu, K. Chen, Q. Tian, W. Wang, and M. D. Plumbley, “AudioSR: Versatile audio super-resolution at scale,” inProc. ICASSP, 2024, pp. 1076–1080

    2024

  2. [10]

    ESPnet: End-to-end speech processing toolkit,

    S. Watanabe, T. Hori, S. Karita, T. Hayashi, J. Nishitoba, Y . Unno, N. E. Y . Soplin, J. Heymann, M. Wiesner, N. Chen, A. Renduch- intala, and T. Ochiai, “ESPnet: End-to-end speech processing toolkit,” inProc. Interspeech, 2018

    2018

  3. [11]

    Summary on the multimodal information-based speech processing 2023 challenge,

    H. Chen, S. Wu, C. Wang, J. Du, C.-H. Lee, S. M. Sinis- calchi, S. Watanabe, J. Chen, O. Scharenborg, Z.-Q. Wang, B.-C. Yin, and J. Pan, “Summary on the multimodal information-based speech processing 2023 challenge,” inProc. ICASSPW, 2024, pp. 123–124

    2023

  4. [12]

    As- teroid: the PyTorch-based audio source separation toolkit for re- searchers,

    M. Pariente, S. Cornell, J. Cosentino, S. Sivasankaran, E. Tzinis, J. Heitkaemper, M. Olvera, F.-R. St ¨oter, M. Hu, J. M. Mart ´ın- Do˜nas, D. Ditter, A. Frank, A. Deleforge, and E. Vincent, “As- teroid: the PyTorch-based audio source separation toolkit for re- searchers,” in...

    2020

  5. [13]

    DeepFilterNet: Perceptually motivated real-time speech en- hancement,

    H. Schr ¨oter, A. N. Escalante, , T. Rosenkranz, and A. Maier, “DeepFilterNet: Perceptually motivated real-time speech en- hancement,” inProc. Interspeech, 2023

    2023

  6. [14]

    Hifi-SR: A unified generative transformer-convolutional adversarial net- work for high-fidelity speech super-resolution,

    S. Zhao, K. Zhou, Z. Pan, Y . Ma, C. Zhang, and B. Ma, “Hifi-SR: A unified generative transformer-convolutional adversarial net- work for high-fidelity speech super-resolution,” inProc. ICASSP, 2025

    2025

  7. [15]

    FlowA VSE: Ef- ficient audio-visual speech enhancement with conditional flow matching,

    C. Jung, S. Lee, J.-H. Kim, and J. S. Chung, “FlowA VSE: Ef- ficient audio-visual speech enhancement with conditional flow matching,” inProc. Interspeech, 2024

    2024

  8. [16]

    FRCRN: Boosting feature representation using frequency recurrence for monaural speech enhancement,

    S. Zhao, B. Ma, K. N. Watcharasupat, and W.-S. Gan, “FRCRN: Boosting feature representation using frequency recurrence for monaural speech enhancement,” inProc. ICASSP, 2022

    2022

  9. [17]

    MossFormer2: Combin- ing transformer and rnn-free recurrent network for enhanced time- domain monaural speech separation,

    S. Zhao, Y . Ma, C. Ni, C. Zhang, H. Wang, T. H. Nguyen, K. Zhou, J. Yip, D. Ng, and B. Ma, “MossFormer2: Combin- ing transformer and rnn-free recurrent network for enhanced time- domain monaural speech separation,” inProc. ICASSP, 2024

    2024

  10. [18]

    MossFormer: Pushing the performance limit of monaural speech separation using gated single-head trans- former with convolution-augmented joint self-attentions,

    S. Zhao and B. Ma, “MossFormer: Pushing the performance limit of monaural speech separation using gated single-head trans- former with convolution-augmented joint self-attentions,” inProc. ICASSP, 2023

    2023

  11. [19]

    Neu- roHeed: Neuro-steered speaker extraction using EEG signals,

    Z. Pan, M. Borsdorf, S. Cai, T. Schultz, and H. Li, “Neu- roHeed: Neuro-steered speaker extraction using EEG signals,” arXiv:2307.14303, 2023

    2023 arXiv

  12. [20]

    HiFi-GAN: Generative adversarial networks for efficient and high fidelity speech synthesis,

    J. Kong, J. Kim, and J. Bae, “HiFi-GAN: Generative adversarial networks for efficient and high fidelity speech synthesis,” inProc. NeurIPS, 2020

    2020

  13. [21]

    Scenario-aware audio-visual TF- Gridnet for target speech extraction,

    Z. Pan, G. Wichern, Y . Masuyama, F. G. Germain, S. Khu- rana, C. Hori, and J. Le Roux, “Scenario-aware audio-visual TF- Gridnet for target speech extraction,” inProc. ASRU, 2023

    2023

  14. [22]

    Speaker extraction with co-speech gestures cue,

    Z. Pan, X. Qian, and H. Li, “Speaker extraction with co-speech gestures cue,”IEEE Signal Process. Lett., vol. 29, pp. 1467–1471, 2022

    2022

  15. [23]

    SpEx+: A complete time domain speaker extraction network,

    M. Ge, C. Xu, L. Wang, E. S. Chng, J. Dang, and H. Li, “SpEx+: A complete time domain speaker extraction network,” inProc. Interspeech, 2020, pp. 1406–1410

    2020

  16. [24]

    DCCRN+: Channel-wise subband dccrn with snr estimation for speech enhancement,

    S. Lv, Y . Hu, S. Zhang, and L. Xie, “DCCRN+: Channel-wise subband dccrn with snr estimation for speech enhancement,” in Proc. Interspeech, 2021

    2021

  17. [25]

    The interspeech 2020 deep noise suppression challenge: Datasets, subjective testing framework, and challenge results,

    C. K. Reddy, V . Gopal, R. Cutler, E. Beyrami, R. Cheng, H. Dubey, S. Matusevych, R. Aichner, A. Aazami, and S. B. et al., “The interspeech 2020 deep noise suppression challenge: Datasets, subjective testing framework, and challenge results,” in Proc. Interspeech, 2020

    2020

  18. [26]

    CSTR VCTK Corpus: English multi-speaker corpus for cstr voice cloning toolkit (version 0.92),

    J. Yamagishi, C. Veaux, and K. MacDonald, “CSTR VCTK Corpus: English multi-speaker corpus for cstr voice cloning toolkit (version 0.92),” 2019. [Online]. Available: https: //api.semanticscholar.org/CorpusID:213060286

    2019

  19. [27]

    Audio set: An ontology and human-labeled dataset for audio events,

    J. F. Gemmeke, D. P. W. Ellis, D. Freedman, A. Jansen, W. Lawrence, R. C. Moore, M. Plakal, and M. Ritter, “Audio set: An ontology and human-labeled dataset for audio events,” inProc. ICASSP, 2017, pp. 776–780

    2017

  20. [28]

    DEMAND: a collection of multi-channel recordings of acoustic noise in diverse environments,

    J. Thiemann, N. Ito, and E. Vincent, “DEMAND: a collection of multi-channel recordings of acoustic noise in diverse environments,” Jun. 2013. [Online]. Available: https://doi.org/10. 5281/zenodo.1227121

    2013

  21. [29]

    Phase- sensitive and recognition-boosted speech separation using deep recurrent neural networks,

    H. Erdogan, J. R. Hershey, S. Watanabe, and J. Le Roux, “Phase- sensitive and recognition-boosted speech separation using deep recurrent neural networks,” inProc. ICASSP, 2015, pp. 708–712

    2015

  22. [30]

    A mask free neural network for monaural speech enhancement,

    L. Liu, H. Guan, J. Ma, W. Dai, G. Wang, and S. Ding, “A mask free neural network for monaural speech enhancement,” inProc. Interspeech, 2023

    2023

  23. [31]

    TridentSE: Guiding speech enhancement with 32 global tokens,

    D. Yin, Z. Zhao, C. Tang, Z. Xiong, and C. Luo, “TridentSE: Guiding speech enhancement with 32 global tokens,” inProc. In- terspeech, 2023

    2023

  24. [32]

    LibriTTS: A corpus derived from librispeech for text- to-speech,

    H. Zen, V . Dang, R. Clark, Y . Zhang, R. J. Weiss, Y . Jia, Z. Chen, and Y . Wu, “LibriTTS: A corpus derived from librispeech for text- to-speech,” inProc. Interspeech, 2019

    2019

  25. [33]

    Explor- ing strategies for training deep neural networks,

    H. Larochelle, Y . Bengio, J. Louradour, and P. Lamblin, “Explor- ing strategies for training deep neural networks,”Journal of Ma- chine Learning Research, vol. 10, no. 1, pp. 1–40, 2009

    2009

  26. [34]

    An efficient encoder-decoder archi- tecture with top-down attention for speech separation,

    K. Li, R. Yang, and X. Hu, “An efficient encoder-decoder archi- tecture with top-down attention for speech separation,” inProc. ICLR, 2023

    2023

  27. [35]

    CMGAN: Conformer-based metric gan for speech enhancement,

    R. Cao, S. Abdulatif, and B. Yang, “CMGAN: Conformer-based metric gan for speech enhancement,” inProc. Interspeech, 2022

    2022

  28. [36]

    Permutation invari- ant training of deep models for speaker-independent multi-talker speech separation,

    D. Yu, M. Kolbæk, Z.-H. Tan, and J. Jensen, “Permutation invari- ant training of deep models for speaker-independent multi-talker speech separation,” inProc. ICASSP, 2017

    2017

  29. [37]

    Dual-Path RNN: Efficient long sequence modeling for time-domain single-channel speech sepa- ration,

    Y . Luo, Z. Chen, and T. Yoshioka, “Dual-Path RNN: Efficient long sequence modeling for time-domain single-channel speech sepa- ration,” inProc. ICASSP, 2020

    2020

  30. [38]

    Attention is all you need in speech separation,

    C. Subakan, M. Ravanelli, S. Cornell, M. Bronzi, and J. Zhong, “Attention is all you need in speech separation,” inProc. ICASSP, 2021

    2021

  31. [39]

    Selective listening by synchronizing speech with lips,

    ——, “Selective listening by synchronizing speech with lips,” IEEE/ACM Trans. Audio, Speech, Lang. Process., vol. 30, pp. 1650–1664, 2022

    2022

  32. [40]

    TF-GridNet: Making time-frequency domain models great again for monaural speaker separation,

    Z.-Q. Wang, S. Cornell, S. Choi, Y . Lee, B.-Y . Kim, and S. Watan- abe, “TF-GridNet: Making time-frequency domain models great again for monaural speaker separation,” inProc. ICASSP, 2023

    2023

  33. [41]

    SPMamba: State-space model is all you need in speech separation,

    K. Li, G. Chen, R. Yang, and X. Hu, “SPMamba: State-space model is all you need in speech separation,”arXiv:2404.02063, 2024

    2024 arXiv

  34. [42]

    Time domain audio visual speech separation,

    J. Wu, Y . Xu, S. Zhang, L. Chen, M. Yu, L. Xie, and D. Yu, “Time domain audio visual speech separation,” inProc. ASRU, 2019

    2019

  35. [43]

    MuSE: Multi-modal target speaker extraction with visual cues,

    Z. Pan, R. Tao, C. Xu, and H. Li, “MuSE: Multi-modal target speaker extraction with visual cues,” inProc. ICASSP, 2021

    2021

  36. [45]

    USEV: Universal speaker extraction with visual cue,

    Z. Pan, M. Ge, and H. Li, “USEV: Universal speaker extraction with visual cue,”IEEE/ACM Trans. Audio, Speech, Lang. Pro- cess., vol. 30, pp. 3032–3045, 2022

    2022

  37. [2021]

    Available: https://arxiv.org/abs/2106.04624

    [Online]. Available: https://arxiv.org/abs/2106.04624

    arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.