pith. sign in

arxiv: 2606.23332 · v1 · pith:QOZI32NUnew · submitted 2026-06-22 · 📡 eess.AS

Don't Listen to Me: A Lightweight, Low-Latency Model for Own-Voice Cancellation in Far-Field Speech Enhancement

Mads {\O}stergaard , Alexander Neergaard Zahid , Karl Ulb{\ae}k , Andreas Hansen Bagge , Kenny Falk{\ae}r Olsen , Rasmus Malik H{\o}egh Lindrup This is my paper

Pith reviewed 2026-06-26 06:47 UTC · model grok-4.3

classification 📡 eess.AS
keywords own-voice cancellationfar-field speech enhancementlow-latency processingspeaker enrollmenttime-domain modelmulti-speaker mixtureMamba architecture
0
0 comments X

The pith

Own-voice cancellation removes an enrolled speaker from far-field mixtures at 2 ms latency while keeping other speech.

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

The paper defines own-voice cancellation as the task of removing only an enrolled target speaker from a noisy multi-speaker mixture. This addresses distortion that occurs when far-field devices send enhanced audio back to the user, because round-trip delay exceeds the threshold for perceiving one's own voice as natural. A time-domain model is conditioned on a short enrollment utterance to identify the target speaker. The authors compare TD-SpeakerBeam to a lighter Mamba-MinGRU masker and show that swapping the auxiliary network for a linear RNN encoder raises signal-to-distortion ratio and predicted MOS while cutting compute. If correct, the result supplies a concrete low-latency objective for real-time far-field denoising.

Core claim

Own-voice cancellation removes a target enrolled speaker from a noisy multi-speaker mixture while preserving any remaining speech. Framed as the complement of target speaker extraction, the task is solved by conditioning a time-domain model on a short enrollment utterance. The model achieves only 2 ms algorithmic latency, and replacing the ConvTasNet-based auxiliary network with a linear RNN encoder improves both signal-to-distortion ratio and predicted MOS while reducing compute.

What carries the argument

Conditioning a time-domain masker on a short enrollment utterance to isolate and cancel only the enrolled speaker.

If this is right

  • OVC functions as a practical low-latency enhancement objective for far-field denoising.
  • The Mamba-MinGRU masker offers a lighter alternative that matches or exceeds TD-SpeakerBeam performance.
  • Linear RNN encoder replacement improves both objective metrics and perceptual quality scores.
  • The approach directly mitigates own-voice artifacts caused by device round-trip latency.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Devices could stream processed audio without the user's voice creating an unnatural echo.
  • The same conditioning technique might extend to selective enhancement of only non-enrolled voices.
  • Real-time implementation on embedded hardware would confirm whether the reported 2 ms latency holds under actual acoustic conditions.

Load-bearing premise

Conditioning on a short enrollment utterance suffices to isolate and remove only the target speaker without harming other speech.

What would settle it

A controlled test in which the model either removes non-enrolled speakers or leaves the enrolled speaker audible after processing.

Figures

Figures reproduced from arXiv: 2606.23332 by Alexander Neergaard Zahid, Andreas Hansen Bagge, Karl Ulb{\ae}k, Kenny Falk{\ae}r Olsen, Mads {\O}stergaard, Rasmus Malik H{\o}egh Lindrup.

Figure 1
Figure 1. Figure 1: Difference between own voice cancellation (OVC) and target speaker extraction (TSE). Given a mixture consisting of multiple speakers recorded in a noisy scene, TSE (right) aims to keep only the enrolled speaker, while OVC (left) removes only the enrolled speaker. Both methods jointly denoise and isolate speakers. global temporal context while supporting causal, streaming in￾ference, making them attractive … view at source ↗
Figure 2
Figure 2. Figure 2: High-level architecture of a time-domain conditioned ConvTasNet. Note that the two encoders do not share parame￾ters. The output of the auxiliary network is an embedding which is applied using an adaptation layer. ity po and the enrolled speaker with probability pe, following [14]. If the other speaker is absent, the target output should cor￾respond to silence, and if the enrolled speaker is absent, the ta… view at source ↗
Figure 3
Figure 3. Figure 3: Detailed architecture of the Mamba-MinGRU masker. It contains an initial normalization layer, followed by a projec￾tion to dmodel, and then N Mamba-MinGRU blocks. A final projection projects the predicted mask back to the encoder di￾mension and applies a non-linearity, here a Sigmoid. 3.3. Auxiliary network and adaptation We investigate two auxiliary networks: (1) a ConvTasNet net￾work with a single repeti… view at source ↗
Figure 4
Figure 4. Figure 4: SDR improvement (dB) for mixtures with multiple in￾terfering speakers. Model IDs in the legend correspond to those shown in [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
read the original abstract

We introduce own-voice cancellation (OVC): removing a target (enrolled) speaker from a noisy multi-speaker mixture while preserving any remaining speech. Framed as the complement of target speaker extraction, OVC addresses latency-induced own-voice artifacts that arise when a far-field device streams enhanced audio back to the user, as the round-trip time easily exceeds the perceptual threshold for own-voice distortion. We condition a time-domain model with only 2 ms algorithmic latency on a short enrollment utterance and benchmark TD-SpeakerBeam alongside a lighter Mamba-MinGRU masker built from Mamba blocks with MinGRU temporal mixing. Replacing the ConvTasNet-based auxiliary network with a linear RNN encoder improves both signal-to-distortion ratio and predicted MOS while reducing compute. Results establish OVC as a practical, low-latency enhancement objective for far-field denoising.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces own-voice cancellation (OVC) as the complement of target-speaker extraction: conditioning a 2 ms latency time-domain masker on a short enrollment utterance to remove only the enrolled speaker from a noisy multi-speaker mixture while preserving all other speech. It benchmarks TD-SpeakerBeam against a lighter Mamba-MinGRU architecture, shows that swapping the auxiliary network for a linear RNN encoder improves SDR and predicted MOS while lowering compute, and concludes that OVC is a practical objective for far-field devices whose round-trip latency would otherwise distort the user's own voice.

Significance. If the selective-cancellation claim is substantiated, the work supplies a concrete, low-latency formulation that directly addresses a perceptual artifact in streaming far-field enhancement. The replacement of the ConvTasNet auxiliary network by a linear RNN and the introduction of the Mamba-MinGRU masker are concrete engineering contributions that could be adopted in resource-constrained devices.

major comments (2)
  1. [Abstract, §3] Abstract and §3 (model description): the central claim requires that the enrollment-conditioned masker nulls only the target speaker without attenuating non-target speech. No separate quantitative evidence (target-speaker energy reduction, non-target SI-SDR, or speaker-specific PESQ) is referenced; aggregate SDR/MOS gains alone do not confirm selectivity, especially given known variance of short-utterance embeddings in far-field overlap.
  2. [§4] §4 (experiments): the manuscript reports SDR and MOS improvements after replacing the auxiliary network, yet supplies neither dataset statistics (number of speakers, enrollment length distribution, SNR range) nor ablation isolating the contribution of the linear RNN versus the Mamba blocks. Without these, the performance delta cannot be attributed to the claimed architectural change.
minor comments (2)
  1. [§3] Notation for the enrollment embedding and its concatenation or similarity mechanism with the masker input should be defined explicitly (e.g., an equation in §3) rather than left to the TD-SpeakerBeam reference.
  2. [Figures] Figure captions and axis labels should state the exact enrollment duration used at test time and whether enrollment is clean or noisy.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below.

read point-by-point responses

  1. Referee: [Abstract, §3] Abstract and §3 (model description): the central claim requires that the enrollment-conditioned masker nulls only the target speaker without attenuating non-target speech. No separate quantitative evidence (target-speaker energy reduction, non-target SI-SDR, or speaker-specific PESQ) is referenced; aggregate SDR/MOS gains alone do not confirm selectivity, especially given known variance of short-utterance embeddings in far-field overlap.

    Authors: We agree that demonstrating selectivity is essential to substantiate the OVC claim. While the aggregate SDR and MOS results are consistent with selective cancellation of the enrolled speaker, they do not directly isolate target-speaker attenuation from non-target preservation. In the revised manuscript we will add target-speaker energy reduction and non-target SI-SDR metrics to provide explicit quantitative support for the selectivity of the enrollment-conditioned masker. revision: yes

  2. Referee: [§4] §4 (experiments): the manuscript reports SDR and MOS improvements after replacing the auxiliary network, yet supplies neither dataset statistics (number of speakers, enrollment length distribution, SNR range) nor ablation isolating the contribution of the linear RNN versus the Mamba blocks. Without these, the performance delta cannot be attributed to the claimed architectural change.

    Authors: We accept that the experimental reporting is incomplete. The revised version will include full dataset statistics (speaker count, enrollment length distribution, SNR ranges) and a dedicated ablation that isolates the linear RNN encoder from the original ConvTasNet auxiliary network as well as the contribution of the Mamba blocks. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical model evaluation without load-bearing derivations

full rationale

The paper introduces own-voice cancellation as a new task and reports empirical results from training and benchmarking time-domain models (TD-SpeakerBeam and Mamba-MinGRU) conditioned on enrollment utterances. No equations, first-principles derivations, fitted parameters presented as predictions, or self-citation chains appear in the abstract or described content. Claims rest on experimental metrics (SDR, MOS) against external benchmarks rather than any reduction to inputs by construction. This is the common case of a self-contained applied ML paper.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Review limited to abstract; no explicit free parameters, axioms, or invented entities are stated. The approach implicitly assumes standard neural conditioning on enrollment utterances works for the new task.

pith-pipeline@v0.9.1-grok · 5723 in / 1098 out tokens · 33473 ms · 2026-06-26T06:47:34.460400+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

40 extracted references · 21 canonical work pages · 7 internal anchors

  1. [1]

    A particularly challenging scenario arises when a far-field device, such as a table-top microphone, captures, enhances, and streams audio back to the user

    Introduction The problem of enhancing speech degraded by environmen- tal noise, interfering speakers, or reverberant effects has been widely studied and remains a common obstacle in many real- world applications such as telecommunications, smart speakers, and conferencing devices. A particularly challenging scenario arises when a far-field device, such as...

  2. [2]

    Don't Listen to Me: A Lightweight, Low-Latency Model for Own-Voice Cancellation in Far-Field Speech Enhancement

    Related work A substantial body of work has focused on developing neu- ral network architectures for speech enhancement and separa- tion [8, 9, 10, 4, 11]. Notable contributions include TasNet [8] and ConvTasNet [9], with ConvTasNet frequently serving as a baseline for evaluating new methods [4]. arXiv:2606.23332v1 [eess.AS] 22 Jun 2026 The perceptual imp...

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  3. [3]

    base" and

    Methods Given an input mixturey=x s +P i̸=s xi +ncontaining a target speakers(the own-voice), other speaker(s)i, and a noise signaln, the goal is to recover ¯y= P i̸=s xi. Following [14], we train the network using at most one other speaker. 3.1. Dataset We train on a dynamically mixed dataset using LibriSpeech [21] in WHAM! noise [22], as in [16] (althou...

  4. [4]

    When comparing task objectives, OVC and TSE appear comparably difficult (cf

    Results and discussion Our main results are shown in Table 1. When comparing task objectives, OVC and TSE appear comparably difficult (cf. (a1) vs. (b1)), with both achieving∼13 dB SDR in the full mixture condition (F). Moving to a causal setting incurs a moderate drop in SDR for both tasks (cf. (a1) vs. (a2) and (b1) vs. (b2)). Replacing the TD-SpeakerBe...

  5. [5]

    Conclusion We have introduced own-voice cancellation as a practical ob- jective for far-field streamed denoising, showing that methods from target speaker extraction can effectively remove an en- rolled speaker from a noisy mixture. The proposed Mamba- MinGRU architecture matches the performance of ConvTasNet- based baselines at a fraction of the compute,...

  6. [6]

    AI tools were used solely for grammar correction

    Generative AI Use Disclosure In accordance with ISCA policy, generative AI tools were not used as co-authors, nor to develop the source code. AI tools were used solely for grammar correction

  7. [7]

    Tolerable hearing aid delays. I. Estimation of limits imposed by the auditory path alone using simulated hearing losses,

    M. A. Stone and B. C. Moore, “Tolerable hearing aid delays. I. Estimation of limits imposed by the auditory path alone using simulated hearing losses,”Ear and Hearing, vol. 20, no. 3, pp. 182–192, 1999

    1999

  8. [8]

    Disturbance caused by varying prop- agation delay in non-occluding hearing aid fittings,

    J. Groth and M. Birkmose, “Disturbance caused by varying prop- agation delay in non-occluding hearing aid fittings,”International Journal of Audiology, vol. 43, no. 10, pp. 594–599, 2004

    2004

  9. [9]

    Tolerable hearing-aid delays: IV. Effects on subjective disturbance during speech production by hearing-impaired subjects,

    M. A. Stone and B. C. J. Moore, “Tolerable hearing-aid delays: IV. Effects on subjective disturbance during speech production by hearing-impaired subjects,”Ear and Hearing, vol. 26, no. 2, pp. 225–235, 2005

    2005

  10. [10]

    Separate and Re- construct: Asymmetric Encoder-Decoder for Speech Separation,

    U.-H. Shin, S. Lee, T. Kim, and H.-M. Park, “Separate and Re- construct: Asymmetric Encoder-Decoder for Speech Separation,” 2024, arXiv:2406.05983 [eess.AS]

    work page arXiv 2024

  11. [11]

    SepMamba: State-space models for speaker separation using Mamba,

    T. H. Avenstrup, B. Elek, I. L. Mádi, A. B. Schin, M. Mørup, B. S. Jensen, and K. F. Olsen, “SepMamba: State-space models for speaker separation using Mamba,” 2024, arXiv:2410.20997 [cs.SD]

    work page arXiv 2024

  12. [12]

    Mamba: Linear-Time Sequence Modeling with Selective State Spaces

    A. Gu and T. Dao, “Mamba: Linear-Time Sequence Modeling with Selective State Spaces,” 2024, arXiv:2312.00752 [cs.LG]

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  13. [13]

    Were RNNs All We Needed?

    L. Feng, F. Tung, M. O. Ahmed, Y . Bengio, and H. Hajimir- sadeghi, “Were RNNs All We Needed?” 2024, arXiv:2410.01201 [cs]

    work page arXiv 2024

  14. [14]

    TaSNet: Time-Domain Audio Separa- tion Network for Real-Time, Single-Channel Speech Separation,

    Y . Luo and N. Mesgarani, “TaSNet: Time-Domain Audio Separa- tion Network for Real-Time, Single-Channel Speech Separation,” in2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Calgary, AB, Canada, 2018, pp. 696–700

    2018

  15. [15]

    Conv-TasNet: Surpassing Ideal Time-Frequency Magni- tude Masking for Speech Separation,

    ——, “Conv-TasNet: Surpassing Ideal Time-Frequency Magni- tude Masking for Speech Separation,”IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 27, no. 8, pp. 1256–1266, 2019

    2019

  16. [16]

    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,” 2020, arXiv:1910.06379 [eess.AS]

    work page arXiv 2020

  17. [17]

    TF-Locoformer: Transformer with Local Modeling by Convolution for Speech Separation and Enhancement,

    K. Saijo, G. Wichern, F. G. Germain, Z. Pan, and J. L. Roux, “TF-Locoformer: Transformer with Local Modeling by Convolution for Speech Separation and Enhancement,” 2024, arXiv:2408.03440 [eess.AS]

    work page arXiv 2024

  18. [18]

    Tolerable hearing aid delays. II. Estimation of limits imposed during speech production,

    M. A. Stone and B. C. J. Moore, “Tolerable hearing aid delays. II. Estimation of limits imposed during speech production,”Ear and Hearing, vol. 23, no. 4, pp. 325–338, 2002

    2002

  19. [19]

    Improving Speaker Discrimination of Target Speech Extraction With Time-Domain Speakerbeam,

    M. Delcroix, T. Ochiai, K. Zmolikova, K. Kinoshita, N. Tawara, T. Nakatani, and S. Araki, “Improving Speaker Discrimination of Target Speech Extraction With Time-Domain Speakerbeam,” in 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Barcelona, Spain, 2020, pp. 691– 695

    2020

  20. [20]

    Listen only to me! How well can target speech ex- traction handle false alarms?

    M. Delcroix, K. Kinoshita, T. Ochiai, K. Zmolikova, H. Sato, and T. Nakatani, “Listen only to me! How well can target speech ex- traction handle false alarms?” 2022, arXiv:2204.04811 [eess.AS]

    work page arXiv 2022

  21. [21]

    TEA-PSE: Tencent-Ethereal-Audio-Lab Personal- ized Speech Enhancement System for ICASSP 2022 DNS Chal- lenge,

    Y . Ju, W. Rao, X. Yan, Y . Fu, S. Lv, L. Cheng, Y . Wang, L. Xie, and S. Shang, “TEA-PSE: Tencent-Ethereal-Audio-Lab Personal- ized Speech Enhancement System for ICASSP 2022 DNS Chal- lenge,” in2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Singapore, Singapore, 2022, pp. 9291–9295

    2022

  22. [22]

    SpeakerBeam-SS: Real-time Target Speaker Extraction with Lightweight Conv-TasNet and State Space Modeling,

    H. Sato, T. Moriya, M. Mimura, S. Horiguchi, T. Ochiai, T. Ashihara, A. Ando, K. Shinayama, and M. Delcroix, “SpeakerBeam-SS: Real-time Target Speaker Extraction with Lightweight Conv-TasNet and State Space Modeling,” 2024, arXiv:2407.01857 [eess.AS]

    work page arXiv 2024

  23. [23]

    Target Speech Extraction with Pre-trained Self- supervised Learning Models,

    J. Peng, M. Delcroix, T. Ochiai, O. Plchot, S. Araki, and J. Cernocky, “Target Speech Extraction with Pre-trained Self- supervised Learning Models,” 2024, arXiv:2402.13199

    work page arXiv 2024

  24. [24]

    Diagonal State Spaces are as Effective as Structured State Spaces,

    A. Gupta, A. Gu, and J. Berant, “Diagonal State Spaces are as Effective as Structured State Spaces,” 2022, arXiv:2203.14343 [cs.LG]

    work page arXiv 2022

  25. [25]

    ICASSP 2023 Acoustic Echo Cancellation Challenge,

    R. Cutler, A. Saabas, T. Pärnamaa, M. Purin, E. Indenbom, N.-C. Ristea, J. Gužvin, H. Gamper, S. Braun, and R. Aichner, “ICASSP 2023 Acoustic Echo Cancellation Challenge,”IEEE Open Journal of Signal Processing, vol. 5, pp. 675–685, 2024

    2023

  26. [26]

    A Progressive Neural Network for Acoustic Echo Can- cellation,

    Z. Chen, X. Xia, S. Sun, Z. Wang, C. Chen, G. Xie, P. Zhang, and Y . Xiao, “A Progressive Neural Network for Acoustic Echo Can- cellation,” in2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Rhodes Island, Greece, 2023, pp. 1–2

    2023

  27. [27]

    Lib- rispeech: An ASR corpus based on public domain audio books,

    V . Panayotov, G. Chen, D. Povey, and S. Khudanpur, “Lib- rispeech: An ASR corpus based on public domain audio books,” in2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), South Brisbane, Queensland, Aus- tralia, 2015, pp. 5206–5210

    2015

  28. [28]

    WHAM!: Extending Speech Separation to Noisy Environments

    G. Wichern, J. Antognini, M. Flynn, L. R. Zhu, E. Mc- Quinn, D. Crow, E. Manilow, and J. L. Roux, “WHAM!: Extending Speech Separation to Noisy Environments,” 2019, arXiv:1907.01160 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  29. [29]

    arXiv preprint arXiv:2005.11262 , year=

    J. Cosentino, M. Pariente, S. Cornell, A. Deleforge, and E. Vin- cent, “LibriMix: An Open-Source Dataset for Generalizable Speech Separation,” 2020, arXiv:2005.11262 [eess]

    work page arXiv 2020

  30. [30]

    Layer Normalization

    J. L. Ba, J. R. Kiros, and G. E. Hinton, “Layer Normalization,” 2016, arXiv:1607.06450 [stat]

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  31. [31]

    Sigmoid-Weighted Linear Units for Neural Network Function Approximation in Reinforcement Learning

    S. Elfwing, E. Uchibe, and K. Doya, “Sigmoid-Weighted Linear Units for Neural Network Function Approximation in Reinforce- ment Learning,” 2017, arXiv:1702.03118 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  32. [32]

    Resurrecting recurrent neural networks for long sequences, 2023

    A. Orvieto, S. L. Smith, A. Gu, A. Fernando, C. Gulcehre, R. Pas- canu, and S. De, “Resurrecting Recurrent Neural Networks for Long Sequences,” 2023, arXiv:2303.06349 [cs]

    work page arXiv 2023

  33. [33]

    Hydra: Bidirectional State Space Models Through Generalized Matrix Mixers,

    S. Hwang, A. Lahoti, T. Dao, and A. Gu, “Hydra: Bidirectional State Space Models Through Generalized Matrix Mixers,” 2024, arXiv:2407.09941 [cs]

    work page arXiv 2024

  34. [34]

    Unsupervised Sound Separation Using Mixture Invariant Training,

    S. Wisdom, E. Tzinis, H. Erdogan, R. J. Weiss, K. Wilson, and J. R. Hershey, “Unsupervised Sound Separation Using Mixture Invariant Training,” inAdvances in Neural Information Process- ing Systems, vol. 33. Vancouver, BC, Canada: Curran Associates, Inc., 2020, pp. 3846–3857

    2020

  35. [35]

    SDR - half-baked or well done?

    J. L. Roux, S. Wisdom, H. Erdogan, and J. R. Hershey, “SDR - half-baked or well done?” 2018, arXiv:1811.02508 [cs.SD]

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  36. [36]

    Distillation and Pruning for Scal- able Self-Supervised Representation-Based Speech Quality As- sessment,

    B. Stahl and H. Gamper, “Distillation and Pruning for Scal- able Self-Supervised Representation-Based Speech Quality As- sessment,” 2025, arXiv:2502.05356 [eess.AS]

    work page arXiv 2025

  37. [37]

    Decoupled Weight Decay Regularization

    I. Loshchilov and F. Hutter, “Decoupled Weight Decay Regular- ization,” 2019, arXiv:1711.05101 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  38. [38]

    Straight to Zero: Why Linearly Decaying the Learning Rate to Zero Works Best for LLMs,

    S. Bergsma, N. Dey, G. Gosal, G. Gray, D. Soboleva, and J. Hes- tness, “Straight to Zero: Why Linearly Decaying the Learning Rate to Zero Works Best for LLMs,” 2025, arXiv:2502.15938 [cs.LG]

    work page arXiv 2025

  39. [39]

    PYIN: A fundamental frequency esti- mator using probabilistic threshold distributions,

    M. Mauch and S. Dixon, “PYIN: A fundamental frequency esti- mator using probabilistic threshold distributions,” in2014 IEEE International Conference on Acoustics, Speech and Signal Pro- cessing (ICASSP), Florence, Italy, 2014, pp. 659–663

    2014

  40. [40]

    YIN, a fundamental fre- quency estimator for speech and music,

    A. De Cheveigné and H. Kawahara, “YIN, a fundamental fre- quency estimator for speech and music,”The Journal of the Acoustical Society of America, vol. 111, no. 4, pp. 1917–1930, 2002

    1917