Pith. sign in

REVIEW 2 major objections 5 minor 38 references

Text-Independent Speaker Verification Using Discrete Audio Tokens

T0 review · 2 major / 5 minor · reviewed 2026-07-14 · grok-4.5

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

Pith's one-line read Speaker identity lives in neural-codec tokens; cross-feature distillation from an Fbank teacher unlocks it for automatic speaker verification.

desk verdict Clean diagnostics show EnCodec tokens keep speaker cues; simple cross-feature embedding KD mostly closes the ASV gap to Fbanks, with a useful architectural probe on the side. read the letter →

arxiv 2607.07579 v2 pith:UGFILF3V submitted 2026-07-08 eess.AS

Zheng Liang , Junjie Li , Kong Aik Lee This is my paper
classification eess.AS
keywords AutomaticSpeakerVerificationNeuralAudioCodecSpeechTokenizerDiscreteRepresentationKnowledgeDistillationEnCross-Feature
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

Neural audio codecs compress speech into compact discrete tokens that work well for generation, yet they lag far behind ordinary filterbank features when used for text-independent speaker verification. The paper shows the shortfall is not mainly missing speaker information: reconstructing the waveform and re-extracting filterbanks recovers almost the original accuracy, proving the cues survive compression. The real problem is that standard classification training cannot dig those cues out of the highly compressed discrete space. By forcing a token-based student network to match the speaker-embedding geometry of a strong filterbank teacher (Cross-Feature Knowledge Distillation), the authors close most of the gap. On VoxCeleb the distilled token systems approach the continuous teacher, and the same pattern holds across bit-rates and architectures. This matters because it opens the door to using the same compact tokens for both generation and recognition, rather than maintaining separate feature pipelines.

What carries the argument

Cross-Feature Knowledge Distillation (CFKD): a frozen Fbank teacher produces target speaker embeddings; a token student (same backbone, input formed by summing residual-vector-quantizer codebook embeddings) is trained with ordinary classification loss plus a weighted cosine-similarity term that aligns its embeddings to the teacher's, thereby transferring speaker geometry across continuous and discrete front-ends.

What would settle it

Train an identical student with the same summed-token input but replace the cosine distillation loss by ordinary classification only, or by a reconstruction loss that forces the student to recover filterbanks; if the large EER reduction disappears, the claim that geometric alignment is the key unlock fails.

Watch

Extended reading notes

Core claim

Speaker-discriminative information is largely preserved inside EnCodec discrete tokens; the performance drop relative to filterbanks arises because conventional ASV training cannot extract it, and embedding-level cosine distillation from an Fbank teacher supplies the missing structured supervision that lets token-based models approach continuous-feature accuracy.

Load-bearing premise

Simply summing residual codebook embeddings and projecting them to the usual filterbank dimension yields a representation whose speaker geometry can be aligned to a continuous teacher by cosine loss alone, without extra terms that model residual quantization noise.

Editorial extensions

If this is right

  • Codec tokens can serve as a shared front-end for both generative and discriminative speech tasks once distillation is applied.
  • 1-D order-agnostic backbones (e.g., ECAPA-TDNN) are preferable to 2-D spectral CNNs for discrete residual-vector-quantizer tokens.
  • Higher distillation weights (around 40) are required for cross-feature transfer than for ordinary same-feature distillation.
  • Performance remains competitive down to a few kilobits per second, suggesting practical low-bitrate speaker verification pipelines.

Reading between the lines

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

  • The complementary error sets between teacher and student imply that a late fusion of continuous and discrete embeddings could exceed either alone.
  • The same distillation recipe may transfer to other recognition tasks (ASR, emotion, language ID) that currently struggle with codec tokens.
  • If residual quantization noise is the residual gap, an explicit noise-aware loss term or multi-codebook attention could close it further without larger teachers.
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

2 major / 5 minor

Summary. The paper argues that discrete tokens from neural audio codecs (e.g., EnCodec RVQ) largely preserve speaker-discriminative information, yet conventional ASV training under-utilizes them, producing a large performance gap relative to continuous Fbank features. Diagnostic experiments (Table 1) separate mild reconstruction loss (E1 o E2) from a larger direct-token gap (E3). The authors introduce Cross-Feature Knowledge Distillation (CFKD): a token-based student is trained with AAM-Softmax plus a cosine-embedding alignment loss (Eqs. 2–4) that forces its speaker embeddings to match those of a frozen Fbank teacher of identical backbone. On VoxCeleb1/2, CFKD recovers most of the gap for both ECAPA-TDNN and ResNet34, across bit-rates, and yields complementary error patterns (Table 4) and architecture-specific inductive-bias evidence (feature-shuffling probe, Table 5).

Significance. If the empirical claims hold, the work supplies a practical and conceptually clean route for making discrete codec tokens competitive with spectral features on a core discriminative task. This matters for the growing ecosystem of neural-codec language models and unified speech systems that already operate on tokens for generation; demonstrating that the same tokens can support high-accuracy ASV without architectural overhaul strengthens the case for token-centric pipelines. Strengths include the clean diagnostic separation of information loss versus utilization loss, consistent gains across two backbones and multiple bit-rates, the error-intersection and shuffling probes that give mechanistic insight, and the use of public VoxCeleb protocols plus a standard toolkit (Wespeaker). The result is therefore both immediately useful and falsifiable.

major comments (2)
  1. [Table 3 / §5.4] Table 3 / §5.4: The headline comparison against Codec-ASV (M5) is confounded. The authors train on VoxCeleb2 alone with a Wespeaker Fbank teacher, while Codec-ASV is reported on Vox1+Vox2; bit-rate points and evaluation protocols also differ slightly. Absolute numbers are still informative, but the claimed “49.5 % relative improvement” and “significant performance improvements across tested bitrates” cannot be attributed solely to CFKD without a matched re-implementation or an explicit statement of the residual protocol mismatch.
  2. [Section 4.1 / Eq. (1)] Section 4.1 and Eq. (1): After summing the 32 RVQ codebook embeddings the authors insert a learnable linear projection to 80 dimensions “to ensure input dimensionality remains consistent.” No ablation is given on (i) whether the projection is necessary, (ii) whether the original EnCodec embedding dimension (typically 128) yields better or worse ASV performance, or (iii) whether a non-linear adapter would further close the residual gap to the teacher. Because the central thesis is that speaker geometry is already present in the tokens and merely needs better supervision, the effect of this dimensionality-reduction step should be quantified; otherwise it remains an unexamined free parameter that could itself be discarding or reshaping speaker cues.
minor comments (5)
  1. [global] Throughout the manuscript (abstract, tables, captions) the dataset name appears as “V oxCeleb” / “V ox1” with an extraneous space; this is a LaTeX artifact that should be corrected for readability.
  2. [Figure 1 / §2] Figure 1 caption and §2: the three diagnostic setups are clearly drawn, yet the main text never states the exact EnCodec configuration (frame rate, codebook size K) used for E3; adding these numbers would make the diagnostic fully self-contained.
  3. [Table 2] Table 2: the relative-improvement column (Δ) is computed only against the naïve student; adding a second column versus the reconstructed-Fbank baseline (E2) would more directly support the claim that CFKD can exceed the information available after decoding.
  4. [§3.2.2] §3.2.2: the notation Φ : X → Z is introduced, yet Z is later used both for the embedding space and for the token matrix; a distinct symbol for the embedding manifold would avoid momentary confusion.
  5. [§1 / Related work] References [14] and [16] are the most directly related prior token-ASV works; a short paragraph contrasting their training recipes (and why CFKD is complementary rather than redundant) would help readers place the contribution.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; claims rest on empirical diagnostics and held-out evaluation, not definitional reduction or self-citation chains.

full rationale

The paper's central argument is empirical rather than a closed-form derivation: Table 1 isolates mild reconstruction loss (E1 2.21 % EER to E2 2.57 %) from the larger direct-token gap (E3 3.38 %), establishing that speaker cues largely survive EnCodec compression. CFKD then supplies embedding-level cosine supervision (Eqs. 2-4) from an Fbank teacher; the resulting student EERs (Table 2 E6 2.25 %, Table 3 across bitrates) are measured on standard VoxCeleb trial lists against an external teacher and against published Codec-ASV numbers. Equation 1 (sum of RVQ embeddings) and the free hyper-parameter lambda do not algebraically force the reported metrics; complementary error sets (Table 4) and the feature-shuffling probe (Table 5) supply independent corroboration. No self-citation is load-bearing for uniqueness or uniqueness theorems, and no fitted constant is renamed as a prediction. The work is therefore self-contained against external benchmarks.

Assumptions & free parameters 3 free parameters · 3 assumptions · 1 invented entities

The work rests on standard ASV and codec assumptions plus a small set of free hyper-parameters (chiefly the distillation weight lambda). No new physical entities are postulated; CFKD is an algorithmic construct whose value is measured by external EER/minDCF.

free parameters (3)
  • lambda (distillation weight) = 40 (best among {0,10,20,40,80})
    Controls the strength of the cosine embedding loss relative to AAM-Softmax; optimal value 40 is selected by grid search on the development set and is substantially larger than typical KD weights.
  • AAM-Softmax margin m and scale s = m=0.2, s=32
    Standard speaker-embedding hyper-parameters fixed at m=0.2, s=32; affect absolute EER but are not claimed as novel.
  • linear projection dimension after token summation = 80
    Maps summed codebook embeddings to 80-D to match Fbank input size; chosen for architectural symmetry rather than derived.
assumptions (3)
  • domain assumption EnCodec residual vector quantizer preserves the majority of speaker-discriminative information at the tested bit-rates.
    Supported by the small E1-to-E2 gap in Table 1; treated as given for the subsequent claim that the remaining gap is utilization, not information loss.
  • domain assumption Cosine similarity of L2-normalized embeddings is a sufficient geometric alignment objective for transferring speaker identity across feature types.
    Invoked in Eqs. 2–3 and Section 3.2.2; standard in modern ASV but not proved optimal for discrete-to-continuous transfer.
  • ad hoc to paper Summing hierarchical RVQ codebook embeddings (Eq. 1) yields a usable dense frame representation for TDNN/ResNet backbones.
    Common practice in cited codec-ASV works, adopted without ablation of alternative fusion methods.
invented entities (1)
  • Cross-Feature Knowledge Distillation (CFKD)
    purpose: Framework that supplies dense embedding-level supervision from an Fbank teacher to a discrete-token student.
    Algorithmic construct introduced in Section 3; its efficacy is measured by external metrics, so independent_evidence is false by definition of the field.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Text-Independent Speaker Verification Using Discrete Audio Tokens." pith.science (2026). https://pith.science/paper/UGFILF3V

@misc{pith2026260707579,
  author       = {Pith},
  title        = {Pith review of: Text-Independent Speaker Verification Using Discrete Audio Tokens},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UGFILF3V}},
  note         = {Machine review of arXiv:2607.07579}
}
read the original abstract

Neural audio codecs (NACs) enable efficient audio compression and have achieved success in downstream tasks such as speech synthesis. However, their discrete representations consistently underperform traditional spectral features in automatic speaker verification (ASV). We empirically demonstrate that speaker cues are implicitly preserved in discrete tokens but remain underutilized by conventional ASV training paradigms. To address this, we propose a Cross-Feature Knowledge Distillation (CFKD) framework. By guiding the codec-based student to mimic the embedding space of a strong Fbank-based teacher, CFKD provides structured supervision for effective utilization of speaker information in tokens. Experiments on the VoxCeleb benchmarks show that CFKD substantially improves the ASV performance of codec-based systems, allowing them to approach the accuracy of Fbank-based teacher models and highlighting the potential of discrete audio tokens for diverse speech tasks.

Figures

Figures reproduced from arXiv: 2607.07579 by the authors.

Figure 1
Figure 1. Schematic of the diagnostic benchmarks. Top (E1): The standard baseline using original Fbanks. Middle (E2): The reconstructed baseline using Fbanks extracted from the de￾coded audio. Bottom (E3): The ASV system trained directly on EnCodec tokens. reconstruct speech timbre, prosody, and other core components of speaker identity, they should be adequate for speaker discrim￾ination. To investigate this discrepancy, we … view at source ↗
Figure 2
Figure 2. Schematic illustration of the proposed CFKD framework. The architecture processes raw audio through dual streams: a continuous Fbank-based teacher and a discrete token-based student. The student input is derived by summing the codebook embeddings from the hierarchical quantizers of the neural codec. R T ×F denote the input Fbank features, where T represents the temporal sequence length and F corresponds to the frequ… view at source ↗

Review history (2 revisions) →

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

38 extracted references · 4 linked inside Pith

  1. [1]

    Introduction Text-independent Automatic Speaker Verification (ASV) aims to authenticate a speaker’s identity regardless of phonetic con- tent, requiring the extraction of robust and speaker-specific traits from unconstrained speech signals [1]. State-of-the-art systems predominantly rely on handcrafted acoustic features, such as Mel-frequency cepstral coe...

  2. [2]

    These systems predominantly rely on con- tinuous spectral features like Fbanks [2, 3, 4, 5]

    From Continuous Spectrograms to Discrete Codec Tokens Modern ASV systems typically function as a sequence-to- vector mapping framework, where a deep neural backbone transforms input acoustic sequences into a discriminative latent speaker space [1]. These systems predominantly rely on con- tinuous spectral features like Fbanks [2, 3, 4, 5]. LetX spec ∈ arX...

    arXiv 2026

  3. [3]

    Methodology 3.1. Motivation: The Information Accessibility Gap Before introducing our proposed framework, we conduct a di- agnostic study to identify the sources of performance degrada- tion in token-based ASV . Specifically, we compare three setups using the same ECAPA-TDNN backbone, as illustrated in Fig- ure 1. The results in Table 1 reveal a clear tre...

  4. [4]

    Audio Frontends and Backends For the spectral baseline, we extract 80-dimensional log Fbanks directly from the original 16 kHz waveforms

    Experimental Settings 4.1. Audio Frontends and Backends For the spectral baseline, we extract 80-dimensional log Fbanks directly from the original 16 kHz waveforms. In contrast, for the token-based systems, we utilize the of- ficial pre-trained 24 kHz EnCodec model [11] as the tokenizer. Consequently, all audio recordings are upsampled to 24 kHz prior to ...

  5. [5]

    Both backends produce 192- dimensional speaker embeddings

    [2] and ResNet34 (C= 32) [23], with 14.65M and 6.63M parameters respectively. Both backends produce 192- dimensional speaker embeddings. 4.2. Training and Evaluation Protocol Unless otherwise stated, models are trained on the V oxCeleb1 development set [24], comprising 1,211 speakers. For these ex- periments, we adopt a standard data augmentation chain in...

  6. [6]

    blind spots

    Experimental Results 5.1. The Effectiveness of Distillation In this section, we evaluate the effectiveness of the proposed CFKD training framework across different backbone architec- tures, including ECAPA-TDNN and ResNet, in terms of EER and minDCF, as reported in Table 2. The results show that, un- der various settings of the distillation weightλ, all m...

  7. [7]

    CFKD suc- cessfully extract speaker identity cues from compressed tokens by leveraging the guidance of a discriminative teacher

    Conclusion We have introduced the Cross-Feature Knowledge Distilla- tion (CFKD) framework to tackle the performance degradation when directly applying neural codec tokens to ASV . CFKD suc- cessfully extract speaker identity cues from compressed tokens by leveraging the guidance of a discriminative teacher. Addi- tionally, we empirically demonstrated that...

  8. [8]

    15228223), and The Hong Kong Polytechnic University, Project ID P0049192

    Acknowledgments This work was supported in part by the Research Grants Council of the Hong Kong SAR (Grant No. 15228223), and The Hong Kong Polytechnic University, Project ID P0049192

Show all 38 references
  1. [9]

    The authors reviewed and edited the output as needed and take full responsibility for the content of the work

    Generative AI Use Disclosure During the preparation of this work, the authors used Genera- tive AI to improve the readability and language quality of the manuscript, as well as for assistance with L ATEX code format- ting. The authors reviewed and edited the output as needed a...

  2. [10]

    Overview of speaker modeling and its applications: From the lens of deep speaker representation learning,

    S. Wang, Z. Chen, K. A. Lee, Y . Qian, and H. Li, “Overview of speaker modeling and its applications: From the lens of deep speaker representation learning,”IEEE/ACM Transactions on Au- dio, Speech, and Language Processing, vol. 32, pp. 4971–4998, 2024

    2024

  3. [11]

    Ecapa-tdnn: Emphasized channel attention, propagation and aggregation in tdnn based speaker verification,

    B. Desplanques, J. Thienpondt, and K. Demuynck, “Ecapa-tdnn: Emphasized channel attention, propagation and aggregation in tdnn based speaker verification,”Interspeech 2020, 2020

    2020

  4. [12]

    X-vectors: Robust dnn embeddings for speaker recognition,

    D. Snyder, D. Garcia-Romero, G. Sell, D. Povey, and S. Khudan- pur, “X-vectors: Robust dnn embeddings for speaker recognition,” in2018 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, 2018, pp. 5329–5333

    2018

  5. [13]

    Cam++: A fast and efficient network for speaker verification using context- aware masking,

    H. Wang, S. Zheng, Y . Chen, L. Cheng, and Q. Chen, “Cam++: A fast and efficient network for speaker verification using context- aware masking,” inProc. Interspeech 2023, 2023, pp. 5301–5305

    2023

  6. [14]

    Golden gemini is all you need: Finding the sweet spots for speaker verification,

    T. Liu, K. A. Lee, Q. Wang, and H. Li, “Golden gemini is all you need: Finding the sweet spots for speaker verification,” IEEE/ACM Transactions on Audio, Speech, and Language Pro- cessing, vol. 32, pp. 2324–2337, 2024

    2024

  7. [15]

    A review of deep learning techniques for speech processing,

    A. Mehrish, N. Majumder, R. Bharadwaj, R. Mihalcea, and S. Po- ria, “A review of deep learning techniques for speech processing,” Information Fusion, vol. 99, p. 101869, 2023

    2023

  8. [16]

    Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups,

    G. Hinton, L. Deng, D. Yu, G. E. Dahl, A.-r. Mohamed, N. Jaitly, A. Senior, V . Vanhoucke, P. Nguyen, T. N. Sainathet al., “Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups,”IEEE Signal processing magazine, vol. 29, no....

    2012

  9. [17]

    The geneva minimalistic acoustic parameter set (gemaps) for voice research and affective computing,

    F. Eyben, K. R. Scherer, B. W. Schuller, J. Sundberg, E. Andr ´e, C. Busso, L. Y . Devillers, J. Epps, P. Laukka, S. S. Narayanan et al., “The geneva minimalistic acoustic parameter set (gemaps) for voice research and affective computing,”IEEE transactions on affective computi...

    2015

  10. [18]

    High-fidelity audio compression with improved rvqgan,

    R. Kumar, P. Seetharaman, A. Luebs, I. Kumar, and K. Kumar, “High-fidelity audio compression with improved rvqgan,”Ad- vances in Neural Information Processing Systems, vol. 36, pp. 27 980–27 993, 2023

    2023

  11. [19]

    Soundstream: An end-to-end neural audio codec,

    N. Zeghidour, A. Luebs, A. Omran, J. Skoglund, and M. Tagliasacchi, “Soundstream: An end-to-end neural audio codec,”IEEE/ACM Transactions on Audio, Speech, and Lan- guage Processing, vol. 30, pp. 495–507, 2021

    2021

  12. [20]

    High fidelity neural audio compression,

    A. D ´efossez, J. Copet, G. Synnaeve, and Y . Adi, “High fidelity neural audio compression,”Transactions on Machine Learning Research

  13. [21]

    Neural codec language models are zero-shot text to speech synthesizers,

    S. Chen, C. Wang, Y . Wu, Z. Zhang, L. Zhou, S. Liu, Z. Chen, Y . Liu, H. Wang, J. Liet al., “Neural codec language models are zero-shot text to speech synthesizers,”IEEE Transactions on Audio, Speech and Language Processing, vol. 33, pp. 705–718, 2025

    2025

  14. [22]

    Audiolm: a language modeling approach to audio gener- ation,

    Z. Borsos, R. Marinier, D. Vincent, E. Kharitonov, O. Pietquin, M. Sharifi, D. Roblek, O. Teboul, D. Grangier, M. Tagliasacchi et al., “Audiolm: a language modeling approach to audio gener- ation,”IEEE/ACM transactions on audio, speech, and language processing, vol. 31, pp. 25...

    2023

  15. [23]

    Codec-asv: Ex- ploring neural audio codec for speaker representation learning,

    Y . Lin, F. Zhang, Y . Gao, S. Zhang, and M. Li, “Codec-asv: Ex- ploring neural audio codec for speaker representation learning,” in ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2025, pp. 1–5

    2025

  16. [24]

    Dasb-discrete audio and speech benchmark,

    P. Mousavi, L. Della Libera, J. Duret, A. Ploujnikov, C. Subakan, and M. Ravanelli, “Dasb-discrete audio and speech benchmark,” arXiv preprint arXiv:2406.14294, 2024

    2024 arXiv

  17. [25]

    Discrete audio representation as an alternative to mel-spectrograms for speaker and speech recognition,

    K. C. Puvvada, N. R. Koluguri, K. Dhawan, J. Balam, and B. Ginsburg, “Discrete audio representation as an alternative to mel-spectrograms for speaker and speech recognition,” inICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). ...

    2024

  18. [26]

    Distilling the knowledge in a neural network,

    G. Hinton, O. Vinyals, and J. Dean, “Distilling the knowledge in a neural network,”arXiv preprint arXiv:1503.02531, 2015

    2015 arXiv

  19. [27]

    Knowledge dis- tillation for small foot-print deep speaker embedding,

    S. Wang, Y . Yang, T. Wang, Y . Qian, and K. Yu, “Knowledge dis- tillation for small foot-print deep speaker embedding,” inICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2019, pp. 6021–6025

    2019

  20. [28]

    IDIR: Identifying and Distilling Informative Relations for Speaker Verification,

    C.-X. Gan, Z. Li, Z. Jin, Z. Huang, M.-W. Mak, and K. A. Lee, “IDIR: Identifying and Distilling Informative Relations for Speaker Verification,” inInterspeech 2025, 2025, pp. 5758–5762

    2025

  21. [29]

    Short utterance compensation in speaker verification via cosine-based teacher- student learning of speaker embeddings,

    J.-w. Jung, H.-S. Heo, H.-j. Shim, and H.-J. Yu, “Short utterance compensation in speaker verification via cosine-based teacher- student learning of speaker embeddings,” in2019 IEEE automatic speech recognition and understanding workshop (ASRU). IEEE, 2019, pp. 335–341

    2019

  22. [30]

    Additive margin softmax for face verification,

    F. Wang, J. Cheng, W. Liu, and H. Liu, “Additive margin softmax for face verification,”IEEE Signal Processing Letters, vol. 25, no. 7, pp. 926–930, 2018

    2018

  23. [31]

    Arcface: Additive angular margin loss for deep face recognition,

    J. Deng, J. Guo, N. Xue, and S. Zafeiriou, “Arcface: Additive angular margin loss for deep face recognition,” inProceedings of the IEEE/CVF conference on computer vision and pattern recog- nition, 2019, pp. 4690–4699

    2019

  24. [32]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 770–778

    2016

  25. [33]

    V oxceleb: a large-scale speaker identification dataset,

    A. Nagrani, J. Chung, and A. Zisserman, “V oxceleb: a large-scale speaker identification dataset,”Interspeech 2017, 2017

    2017

  26. [34]

    Audio augmen- tation for speech recognition

    T. Ko, V . Peddinti, D. Povey, and S. Khudanpur, “Audio augmen- tation for speech recognition.” inInterspeech, vol. 2015, 2015, p. 3586

    2015

  27. [35]

    Musan: A music, speech, and noise corpus,

    D. Snyder, G. Chen, and D. Povey, “Musan: A music, speech, and noise corpus,”arXiv preprint arXiv:1510.08484, 2015

    2015 arXiv

  28. [36]

    A study on data augmentation of reverberant speech for robust speech recognition,

    T. Ko, V . Peddinti, D. Povey, M. L. Seltzer, and S. Khudanpur, “A study on data augmentation of reverberant speech for robust speech recognition,” in2017 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, 2017, pp. 5220–5224

    2017

  29. [37]

    V oxceleb2: Deep speaker recognition,

    J. S. Chung, A. Nagrani, and A. Zisserman, “V oxceleb2: Deep speaker recognition,”Interspeech 2018, 2018

    2018

  30. [38]

    Wespeaker: A research and production oriented speaker embedding learning toolkit,

    H. Wang, C. Liang, S. Wang, Z. Chen, B. Zhang, X. Xiang, Y . Deng, and Y . Qian, “Wespeaker: A research and production oriented speaker embedding learning toolkit,” inICASSP 2023- 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 202...

    2023

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.