Pith. sign in

REVIEW 4 major objections 5 minor 49 references

BUT System for the MLC-SLM Challenge

T0 review · 4 major / 5 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 This paper shows that a Whisper ASR model conditioned on frame-level diarization probabilities, rather than speaker embeddings, transcribes two-speaker conversations in 15 languages and ranks second in the MLC-SLM challenge with a 16.75%…

desk verdict A competent challenge write-up whose real value is the labeling-inconsistency analysis and VAD mitigation, but the VAD case rests on an under-documented proxy and a single test evaluation. read the letter →

arxiv 2506.13414 v1 pith:SA7LEACK submitted 2025-06-16 eess.AS

Alexander Polok , Jiangyu Han , Dominik Klement , Samuele Cornell , Jan Černocký , Lukáš Burget This is my paper · ORCID
classification eess.AS
keywords diarization-conditionedASRtarget-speakerspeakerdiarizationWhisperWavLMvoiceactivitydetectionmultilingualMLC-SLMchallenge
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 describes a complete two-speaker ASR pipeline for the MLC-SLM challenge: DiariZen, a WavLM-and-Conformer diarization pipeline built on Pyannote, feeds speaker-activity probabilities to DiCoW, a Whisper variant whose encoder applies frame-dependent affine transformations conditioned on four speaking states. The authors claim this combination generalizes across 15 languages without domain adaptation, keeps Whisper's multilingual ability after English-only fine-tuning, and reaches a micro-average tcpWER/CER of 16.75 percent, ranking second in Task 2. The paper also argues that the challenge training data contains labeling inconsistencies, skipped speech and silence marked as speech, that poison diarization fine-tuning, and that adding a Silero voice-activity detector with weight 0.8 improves the test-set tcpWER from 28.6 percent to 17.4 percent. A sympathetic reader would care because the result suggests modular, non-LLM systems can compete on multilingual conversational ASR, and because the labeling-mismatch analysis offers a cautionary lesson for any diarization fine-tuning on loosely annotated data.

What carries the argument

The load-bearing mechanism is the Silence-Target-NonTarget-Overlap (STNO) mask: from diarization probabilities $d(s,t)$ the system forms four per-frame probabilities $p_S, p_T, p_N, p_O$ that sum to one, and each Whisper encoder layer applies four affine transformations $W_l z_t + b_l$, one per state, blended by those probabilities via the FDDT (frame-level diarization-dependent transformation). This carries the conditioning signal from diarization into the ASR encoder without speaker embeddings or enrollment audio. On the diarization side, DiariZen's local EEND module combines a WavLM backbone with a Conformer and a powerset classification head, then clusters speaker embeddings across chunks. The third piece is the auxiliary Silero VAD merged into the diarization probabilities with weight 0.8, which suppresses the false speech predictions the fine-tuned model learned from inconsistently labeled silence.

What would settle it

Re-run DiCoW+DiariZen on the official MLC-SLM test set with the Silero VAD weight set to 0.5, 0.7, 0.9, and 1.0 while keeping everything else fixed; if 0.8 is not near-optimal, or if the gap between fine-tuned-with-VAD and fine-tuned-only narrows below the reported 28.6%-to-17.4% margin, the mitigation claim fails. Independently, have human annotators label a sample of the test set and compute DER for the fine-tuned DiariZen with and without VAD; a large false-alarm rate on human-verified silence would confirm the label-inconsistency story, while low false alarms would contradict it.

Watch

Extended reading notes

Core claim

The central claim is that frame-level diarization masks are a sufficient conditioning signal for target-speaker ASR: DiCoW replaces speaker embeddings with per-frame probabilities of silence, target speech, non-target speech, and overlap, and applies these as a weighted blend of four learned affine transforms in every Whisper encoder layer. Because the transforms are initialized to the identity, the pretrained Whisper's behavior is preserved at the start of fine-tuning, which the authors say explains why the model keeps its multilingual performance even though the fine-tuning data is English-only. On the MLC-SLM Task 2 evaluation, the full system with fine-tuned DiariZen segmentation achieves 16.75% micro-average tcpWER/CER, second place, and DiariZen alone beats the Pyannote baseline on both out-of-domain and fine-tuned conditions. A secondary claim is that apparent DER improvements from fine-tuning can be artifacts: the fine-tuned diarizer learns to mimic the inconsistent training labels, and a VAD-based speech/silence correction improves the test tcpWER from 28.6% to 17.4% even though it makes DER look worse on the dev set.

Load-bearing premise

The claim that the VAD mitigation improves the test set from 28.6% to 17.4% rests on two premises: that the test set follows a cleaner protocol in which all unannotated segments are truly muted, and that the authors' modified 'test-like' development set, where unannotated segments with noticeable energy are muted using an unspecified energy threshold, faithfully reproduces that protocol, so the hand-set Silero VAD weight of 0.8 transfers.

Editorial extensions

If this is right

  • A two-speaker ASR system can be built modularly from diarization and a conditioned Whisper, with no LLM rescoring, and still place second in a 15-language challenge.
  • Target-speaker ASR does not require speaker embeddings or enrollment: frame-level activity probabilities carry enough speaker identity information for Whisper to separate two speakers.
  • Fine-tuning Whisper's encoder with identity-initialized affine transforms on English data does not erase multilingual recognition, so diarization-conditioned adaptation is a viable route to low-resource language support.
  • Diarization fine-tuning on loosely annotated data can hurt downstream ASR by teaching the model to reproduce missing-speech and silence-as-speech labels; auxiliary VAD post-processing is a cheap corrective.
  • Because the test protocol mutes unannotated segments, diarization systems should be evaluated on a test-like protocol with muted unannotated audio, not on the raw dev set, to predict real performance.

Reading between the lines

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

  • The STNO conditioning idea generalizes: the same four-state mask could be fed to a speech LLM decoder in place of Whisper's encoder, which the paper names as future work, and could extend to more than two speakers by partitioning the overlap state.
  • The VAD-weight transfer is the fragile link: 0.8 was chosen without a documented sweep on the modified dev set, so the 17.4% test number should be read as conditional on that hand-set value holding on the true test protocol.
  • The label-inconsistency analysis implies that other challenge participants who fine-tuned diarization on the same data likely absorbed the same bias, so part of the ranking gap may reflect who corrected for annotation artifacts rather than intrinsic diarization quality.
  • A testable extension: recompute the modified 'test-like' dev set with different energy thresholds and check whether the FT+VAD advantage over FT is stable; if it flips, the proxy is not faithful enough to guide system choices.
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 / 5 minor

Summary. This paper describes the BUT system for the MLC-SLM Challenge Task 2: a two-speaker ASR pipeline combining DiariZen (a Pyannote-based diarization pipeline with WavLM/Conformer EEND) with DiCoW (a diarization-conditioned Whisper variant). The authors report that DiariZen outperforms the Pyannote baseline in both out-of-domain and fine-tuned settings on the development data, that DiCoW retains multilingual ASR ability despite English-only target-speaker fine-tuning, and that the final system reaches a micro-average tcpWER/CER of 16.75% and ranks second in the challenge. They also document labeling inconsistencies in the MLC-SLM training/development data (unannotated speech and silence labeled as speech) and propose a Silero VAD-based mitigation, reporting an improvement on the test set from 28.6% to 17.4%.

Significance. If validated, the paper's contribution is a competitive, non-LLM modular system for multilingual two-speaker transcription, plus a useful analysis of diarization training under loosely annotated data. The release of the DiCoW and DiariZen models is a concrete strength, as is the authors' explicit caution (Section 4.1) that DER gains may reflect learning to mimic inconsistent annotations rather than improved diarization. The main concerns are the unexplained discrepancy between the two reported test-set numbers (16.75% vs. 17.4%) and the lack of validation of the key parameters in the VAD mitigation, which is the basis of the system's largest reported improvement.

major comments (4)
  1. [Section 4.3 (test-set results)] The abstract states the final system achieves a micro-average tcpWER/CER of 16.75%, but Section 4.3 reports that the VAD-based approach improves the test set from 28.6% to 17.4%. The relationship between these two test-set numbers is not explained; the paper must clarify which configuration yields 16.75% (e.g., whether 17.4% is an intermediate result, a different evaluation protocol, or a subset) and why they differ.
  2. [Section 4.3 (test-like development set)] The 'test-like' development set is constructed by muting unannotated segments with 'noticeable energy,' but the energy threshold is not specified, and no evidence is provided that this proxy reproduces the official test protocol's 'cleaner' annotation. Since the test-set improvement (28.6% to 17.4%) is the main evidence for the VAD mitigation, the paper should specify the threshold and, ideally, validate the proxy against the actual test protocol or an additional held-out set.
  3. [Section 4.3 (Silero VAD weight)] The Silero VAD weight of 0.8 is introduced without an ablation or tuning curve. This is particularly important because Table 3 shows that FT+VAD worsens DER on the test-like development set (16.2% vs. 12.4% for FT), and the benefit only appears in the downstream tcpWER/CER. To support the transferability of this hand-set weight from the proxy to the test distribution, the paper should report sensitivity to this weight.
  4. [Section 4.2, Table 2] The 'Real diar' comparison confounds two changes: the diarization module (DiariZen vs. Pyannote) and the ASR/inference module (DiCoW with long-form decoding vs. Whisper large-v3 with chunked decoding). Consequently, the reported gains cannot be attributed to the diarization replacement alone. A decomposition (e.g., Pyannote segmentation with DiCoW, or DiariZen segmentation with baseline Whisper) or an explicit statement of this confound is needed.
minor comments (5)
  1. [Table 1] The language label 'Inidian English' should be 'Indian English.'
  2. [Section 4.1] The caution about DER gains being potentially misleading is welcome, but the abstract should also mention this caveat, since it currently states the fine-tuned DiariZen 'continues to outperform the fine-tuned Pyannote baseline' without noting that this may reflect annotation mimicry.
  3. [Table 3] The systems labeled 'FT+V AD' contain a stray space in the 'V AD' acronym; it should be 'FT+VAD' throughout.
  4. [Figure 3] The caption and text do not specify which columns correspond to the four system variants; a short legend or description would improve readability.
  5. [Section 3.2] The sentence 'We use a challenge text normalization procedure to the reference transcripts' is slightly ambiguous; it should say 'apply ... to' rather than 'to.'

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the core results are external benchmark measurements, and no derivation reduces to its inputs.

full rationale

The paper's central claims—DiariZen DER improvements over Pyannote, DiCoW ASR gains, the final 16.75% tcpWER/CER and second place in Task 2—are empirical evaluations against the MLC-SLM challenge's official development/test data and baseline system, not derivations from fitted quantities. DiCoW and DiariZen are prior work by the same group, and the paper cites those papers for the architectures (Sections 2.1, 2.2, 3.1, 3.2), but the benchmark results do not depend on those citations for their validity; the models are released and scored externally. Equation (2) defines the conditioning transform from diarization probabilities, and the STNO probabilities in Eq. (1) are computed from the diarization output—this is a system definition, not a circular derivation, since the target transcript is not used to define the conditioning signal. The Silero VAD weight 0.8 in Section 4.3 and the "test-like" development set constructed by muting unannotated segments are unablated and the proxy-to-test transfer is an assumption, but this is a methodological/correctness risk, not a circularity: the test-set tcpWER gain (28.6% to 17.4%) is an official evaluation outcome, not the value of a fitted parameter renamed as a prediction. The paper itself flags that DER gains may be misleading (Section 4.1), which further shows awareness rather than a circular argument. No self-definitional, fitted-input-as-prediction, or self-citation-load-bearing step is exhibited.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claims rest on the reliability of the MLC-SLM challenge evaluation, the validity of the test-like development proxy, and the transfer of the hand-set VAD weight. The paper's own caveats about misleading DER are incorporated as explicit assumptions rather than hidden ones.

free parameters (2)
  • Silero VAD weight = 0.8
    Hand-set weight for merging VAD probabilities with diarization probabilities; reported without ablation or search procedure, yet it drives the main test-set improvement (28.6% to 17.4%).
  • Energy threshold for unannotated segments in test-like dev set = unspecified
    The modified development set mutes 'unannotated segments with noticeable energy'; the threshold is not quantified, so the proxy is not reproducible.
assumptions (4)
  • domain assumption The MLC-SLM test set protocol mutes unannotated segments, while training and development data contain missing speech and mislabeled silence.
    Section 4.3; the entire VAD mitigation strategy assumes this asymmetry between test and development partitions.
  • domain assumption Diarization DER computed without a forgiveness collar is a meaningful indicator of diarization quality.
    Section 4.1 uses DER without collar; the authors themselves caution that DER gains can be misleading under inconsistent labels.
  • domain assumption Listening to audio confirms that the out-of-domain diarization follows actual speech activity better than ground-truth labels.
    Section 4.3, Figure 3; informal perceptual verification is used as evidence.
  • domain assumption Whisper large-v3-turbo with long-form inference is comparable to or better than large-v3 with chunked inference for these languages, aside from the noted Thai exception.
    Section 4.2; the GT-segmentation comparison between baseline and DiCoW conflates model version and inference mode.

how reviews work

0 comments
Cite this review

Pith. "Pith review of BUT System for the MLC-SLM Challenge." pith.science (2026). https://pith.science/paper/SA7LEACK

@misc{pith2026250613414,
  author       = {Pith},
  title        = {Pith review of: BUT System for the MLC-SLM Challenge},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SA7LEACK}},
  note         = {Machine review of arXiv:2506.13414}
}
read the original abstract

We present a two-speaker automatic speech recognition (ASR) system that combines DiCoW -- a diarization-conditioned variant of Whisper -- with DiariZen, a diarization pipeline built on top of Pyannote. We first evaluate both systems in out-of-domain (OOD) multilingual scenarios without any fine-tuning. In this scenario, DiariZen consistently outperforms the baseline Pyannote diarization model, demonstrating strong generalization. Despite being fine-tuned on English-only data for target-speaker ASR, DiCoW retains solid multilingual performance, indicating that encoder modifications preserve Whisper's multilingual capabilities. We then fine-tune both DiCoW and DiariZen on the MLC-SLM challenge data. The fine-tuned DiariZen continues to outperform the fine-tuned Pyannote baseline, while DiCoW sees further gains from domain adaptation. Our final system achieves a micro-average tcpWER/CER of 16.75% and ranks second in Task 2 of the MLC-SLM challenge. Lastly, we identify several labeling inconsistencies in the training data -- such as missing speech segments and incorrect silence annotations -- which can hinder diarization fine-tuning. We propose simple mitigation strategies to address these issues and improve system robustness.

Figures

Figures reproduced from arXiv: 2506.13414 by the authors.

Figure 1
Figure 1. Framework of local EEND module for DiariZen. Fig￾ure adapted from [26] [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of the DiCoW model architecture. The model is based on the Whisper architecture, with modifications to incorporate frame-level diarization information through Frame-Level Diarization Dependent Transformations (FDDT). Figure adapted from [24]. 2.1. DiariZen DiariZen is a speaker diarization pipeline built on Pyannote. Given a long audio recording, DiariZen first segments the in￾put into shorter chunks and th… view at source ↗
Figure 3
Figure 3. shows an example from the development set (English-American-0525 002). The difference between the spectrogram and the GT segmentation reveals skipped speech and inaccurate overlaps. On the other hand, the OOD system closely follows actual speech activity, which was confirmed by listening to the audio, while fine-tuning shifts predictions to￾ward inconsistent GT labels. Although FT+VAD seemingly worsens DER, it effec… view at source ↗

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 44 canonical work pages

  1. [1]

    These models achieve remarkable accuracy by leveraging mas- sive training data [7, 8] and scaling up model parameters [9]

    Introduction Recent advances in deep learning, particularly the rise of large self-supervised models [1, 2], large language mod- els (LLMs) [3, 4], and Whisper-style supervised architec- tures [5, 6], have led to substantial progress in automatic speech recognition (ASR), even under challenging acoustic conditions. These models achieve remarkable accuracy...

    work page

  2. [2]

    BUT System for the MLC-SLM Challenge

    Method This section provides an overview of our system, which is de- signed similarly to the official baseline3 but with two key differ- ences: the Pyannote diarization module is replaced by DiariZen, and unlike the baseline system—which processes each chunk extracted from diarization independently, either with Whisper or Whisper connected to an LLM—our a...

    work page Pith review arXiv 2025

  3. [3]

    DiariZen Our diarization system builds upon the DiariZen framework 4, following the training approach described in [26]

    Experimental Setup 3.1. DiariZen Our diarization system builds upon the DiariZen framework 4, following the training approach described in [26]. We use WavLM Large as the backbone for frame-level classification to improve local modeling. The model is pre-trained on far- field, single-channel audio from a diverse collection of pub- lic datasets, including ...

    work page

  4. [4]

    test-like

    Results In this section, we evaluate our system’s performance. We be- gin with diarization error rates (DERs), followed by TS-ASR results using both ground truth and DiariZen-derived segmenta- tions. We then analyze labeling inconsistencies in the training and development data and their impact, and to address these is- sues, we incorporate an auxiliary V ...

    work page

  5. [5]

    Our system placed second in the MLC-SLM Challenge, showing strong performance across diverse lan- guages

    Conclusions We presented a simple, non-LLM approach for multilin- gual long-form transcription combining diarization and target- speaker ASR. Our system placed second in the MLC-SLM Challenge, showing strong performance across diverse lan- guages. While effective, our method has limitations that open avenues for future work. In particular, we did not full...

    work page

  6. [6]

    Linguistics, Artificial Intelligence and Language and Speech Technologies: from Research to Applications

    Acknowledgements The work was supported by Ministry of Education, Youth and Sports of the Czech Republic (MoE) through the OP JAK project “Linguistics, Artificial Intelligence and Language and Speech Technologies: from Research to Applications” (ID:CZ.02.01.01/00/23 020/0008518) and by Czech Ministry of Interior project No. VK01020132 “112”. Computing on ...

    work page

  7. [7]

    WavLM: Large-scale self-supervised pre-training for full stack speech processing,

    S. Chenet al., “WavLM: Large-scale self-supervised pre-training for full stack speech processing,”IEEE Journal of Selected Topics in Signal Processing, 2022

    work page 2022

  8. [8]

    HuBERT: How much can a bad teacher benefit ASR pre-training?

    W.-N. Hsuet al., “HuBERT: How much can a bad teacher benefit ASR pre-training?” in2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2021

    work page 2021

Show all 49 references
  1. [9]

    GPT-4 technical report,

    J. Achiamet al., “GPT-4 technical report,”arXiv preprint arXiv:2303.08774, 2023

    2023 arXiv

  2. [10]

    Llama: Open and efficient foundation lan- guage models,

    H. Touvronet al., “Llama: Open and efficient foundation lan- guage models,”arXiv preprint arXiv:2302.13971, 2023

    2023 arXiv

  3. [11]

    Robust speech recognition via large-scale weak supervision,

    A. Radfordet al., “Robust speech recognition via large-scale weak supervision,” inInternational conference on machine learning. PMLR, 2023

    2023

  4. [12]

    Reproducing Whisper-style training using an open-source toolkit and publicly available data,

    Y . Penget al., “Reproducing Whisper-style training using an open-source toolkit and publicly available data,” in2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU), 2023

    2023

  5. [13]

    YODAS: Youtube-oriented dataset for audio and speech,

    X. Liet al., “YODAS: Youtube-oriented dataset for audio and speech,” in2023 IEEE Automatic Speech Recognition and Un- derstanding Workshop (ASRU). IEEE, 2023

    2023

  6. [14]

    GigaSpeech: An evolving, multi-domain ASR corpus with 10,000 hours of transcribed audio,

    G. Chenet al., “GigaSpeech: An evolving, multi-domain ASR corpus with 10,000 hours of transcribed audio,” inInterspeech 2021, 2021

    2021

  7. [15]

    OWLS: Scaling laws for multilingual speech recognition and translation models,

    W. Chenet al., “OWLS: Scaling laws for multilingual speech recognition and translation models,” 2025

    2025

  8. [16]

    The CHiME-7 DASR challenge: Distant meet- ing transcription with multiple devices in diverse scenarios,

    S. Cornellet al., “The CHiME-7 DASR challenge: Distant meet- ing transcription with multiple devices in diverse scenarios,” in 7th International Workshop on Speech Processing in Everyday Environments (CHiME 2023), 2023

    2023

  9. [17]

    The CHiME-8 DASR challenge for generalizable and ar- ray agnostic distant automatic speech recognition and diariza- tion,

    ——, “The CHiME-8 DASR challenge for generalizable and ar- ray agnostic distant automatic speech recognition and diariza- tion,” in8th International Workshop on Speech Processing in Ev- eryday Environments (CHiME 2024), 2024

    2024

  10. [19]

    The USTC-NERCSLIP systems for the CHiME- 8 NOTSOFAR-1 challenge,

    S. Niuet al., “The USTC-NERCSLIP systems for the CHiME- 8 NOTSOFAR-1 challenge,” in8th International Workshop on Speech Processing in Everyday Environments (CHiME 2024), 2024

    2024

  11. [20]

    BUT/JHU system description for CHiME- 8 NOTSOFAR-1 challenge,

    A. Poloket al., “BUT/JHU system description for CHiME- 8 NOTSOFAR-1 challenge,” in8th International Workshop on Speech Processing in Everyday Environments (CHiME 2024), 2024

    2024

  12. [21]

    The NPU-TEA system for the CHiME-8 NOTSOFAR-1 challenge,

    K. Huanget al., “The NPU-TEA system for the CHiME-8 NOTSOFAR-1 challenge,” in8th International Workshop on Speech Processing in Everyday Environments (CHiME 2024), 2024

    2024

  13. [22]

    Serialized output training for end-to-end over- lapped speech recognition,

    N. Kandaet al., “Serialized output training for end-to-end over- lapped speech recognition,” inInterspeech 2020, 2020

    2020

  14. [23]

    One model to rule them all? Towards end-to-end joint speaker diarization and speech recognition,

    S. Cornellet al., “One model to rule them all? Towards end-to-end joint speaker diarization and speech recognition,” in2024 IEEE International Conference on Acoustics, Speech and Signal Pro- cessing (ICASSP). IEEE, 2024

    2024

  15. [24]

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

    D. Yuet al., “Permutation invariant training of deep models for speaker-independent multi-talker speech separation,” in2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2017

    2017

  16. [25]

    Auxiliary interference speaker loss for target- speaker speech recognition,

    N. Kandaet al., “Auxiliary interference speaker loss for target- speaker speech recognition,” inInterspeech 2019, 2019

    2019

  17. [26]

    Conformer-based target-speaker automatic speech recognition for single-channel audio,

    Y . Zhanget al., “Conformer-based target-speaker automatic speech recognition for single-channel audio,” in2023 IEEE Inter- national Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023

    2023

  18. [27]

    Adapting self-supervised models to multi-talker speech recognition using speaker embeddings,

    Z. Huanget al., “Adapting self-supervised models to multi-talker speech recognition using speaker embeddings,” in2023 IEEE In- ternational Conference on Acoustics, Speech and Signal Process- ing (ICASSP), 2023

    2023

  19. [28]

    Empowering Whisper as a joint multi-talker and target-talker speech recognition system,

    L. Menget al., “Empowering Whisper as a joint multi-talker and target-talker speech recognition system,” inInterspeech 2024, 2024

    2024

  20. [29]

    Target speaker ASR with Whisper,

    A. Poloket al., “Target speaker ASR with Whisper,” in2025 IEEE International Conference on Acoustics, Speech and Signal Pro- cessing (ICASSP), 2025

    2025

  21. [30]

    DiCoW: Diarization-conditioned Whisper for target speaker automatic speech recognition,

    ——, “DiCoW: Diarization-conditioned Whisper for target speaker automatic speech recognition,” 2024

    2024

  22. [31]

    Mamba-based segmentation model for speaker diarization,

    A. Plaquetet al., “Mamba-based segmentation model for speaker diarization,” inICASSP 2025 - 2025 IEEE International Confer- ence on Acoustics, Speech and Signal Processing (ICASSP), 2025

    2025

  23. [32]

    Leveraging self-supervised learning for speaker di- arization,

    J. Hanet al., “Leveraging self-supervised learning for speaker di- arization,” in2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2025

    2025

  24. [33]

    pyannote.audio 2.1 speaker diarization pipeline: prin- ciple, benchmark, and recipe,

    H. Bredin, “pyannote.audio 2.1 speaker diarization pipeline: prin- ciple, benchmark, and recipe,” inInterspeech 2023, 2023

    2023

  25. [34]

    Powerset multi-class cross entropy loss for neural speaker diarization,

    A. Plaquet and H. Bredin, “Powerset multi-class cross entropy loss for neural speaker diarization,” inInterspeech 2023, 2023

    2023

  26. [35]

    Conformer: Convolution-augmented transformer for speech recognition,

    A. Gulatiet al., “Conformer: Convolution-augmented transformer for speech recognition,” inInterspeech 2020, 2020

    2020

  27. [36]

    The AMI meeting corpus,

    I. Mccowanet al., “The AMI meeting corpus,”Int’l. Conf. on Methods and Techniques in Behavioral Research, 01 2005

    2005

  28. [37]

    AISHELL-4: An open source dataset for speech enhancement, separation, recognition and speaker diarization in conference scenario,

    Y . Fuet al., “AISHELL-4: An open source dataset for speech enhancement, separation, recognition and speaker diarization in conference scenario,” inProc. Interspeech, 2021

    2021

  29. [38]

    M2MeT: The ICASSP 2022 multi-channel multi- party meeting transcription challenge,

    F. Yuet al., “M2MeT: The ICASSP 2022 multi-channel multi- party meeting transcription challenge,” inProc. ICASSP. IEEE, 2022

    2022

  30. [39]

    NOTSOFAR-1 challenge: New datasets, baseline, and tasks for distant meeting transcription,

    A. Vinnikovet al., “NOTSOFAR-1 challenge: New datasets, baseline, and tasks for distant meeting transcription,” inInter- speech 2024, 2024

    2024

  31. [40]

    MSDWild: Multi-modal speaker diarization dataset in the wild

    T. Liuet al., “MSDWild: Multi-modal speaker diarization dataset in the wild.” inINTERSPEECH, 2022

    2022

  32. [41]

    The third DIHARD diarization challenge,

    N. Ryantet al., “The third DIHARD diarization challenge,” in Interspeech 2021, 2021

    2021

  33. [42]

    Open source MagicData-RAMC: A rich annotated Mandarin conversational (RAMC) speech dataset,

    Z. Yanget al., “Open source MagicData-RAMC: A rich annotated Mandarin conversational (RAMC) speech dataset,”arXiv preprint arXiv:2203.16844, 2022

    2022 arXiv

  34. [43]

    Spot the conversation: Speaker diarisation in the wild,

    J. S. Chunget al., “Spot the conversation: Speaker diarisation in the wild,” inInterspeech 2020, 2020

    2020

  35. [44]

    Fine-tune before structured pruning: Towards com- pact and accurate self-supervised models for speaker diarization,

    J. Hanet al., “Fine-tune before structured pruning: Towards com- pact and accurate self-supervised models for speaker diarization,” arXiv preprint arXiv:2505.24111, 2025

    2025 arXiv

  36. [45]

    Advancing speaker embedding learning: Wes- peaker toolkit for research and production,

    S. Wanget al., “Advancing speaker embedding learning: Wes- peaker toolkit for research and production,”Speech Communica- tion, 2024

    2024

  37. [46]

    V oxCeleb2: Deep speaker recognition,

    J. S. Chunget al., “V oxCeleb2: Deep speaker recognition,” in Interspeech 2018, 2018

    2018

  38. [47]

    Librispeech: An ASR corpus based on public domain audio books,

    V . Panayotovet al., “Librispeech: An ASR corpus based on public domain audio books,” in2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2015

    2015

  39. [48]

    LibriMix: An open-source dataset for generalizable speech separation,

    J. Cosentinoet al., “LibriMix: An open-source dataset for generalizable speech separation,”arXiv: Audio and Speech Processing, 2020

    2020

  40. [49]

    Joint CTC/attention decoding for end-to-end speech recognition,

    T. Horiet al., “Joint CTC/attention decoding for end-to-end speech recognition,” inProceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), R. Barzilay and M.-Y . Kan, Eds. Vancouver, Canada: Association for Computational...

    2017

  41. [50]

    Silero V AD: pre-trained enterprise-grade voice activity detector (V AD), number detector and language classifier,

    S. Team, “Silero V AD: pre-trained enterprise-grade voice activity detector (V AD), number detector and language classifier,” 2024

    2024

Pith tools

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