REVIEW 4 major objections 5 minor 45 references
CabinSep: IR-Augmented Mask-Based MVDR for Real-Time In-Car Speech Separation with Distributed Heterogeneous Arrays
T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash
Pith's one-line read A lightweight mask-based MVDR front end cuts in-car speech-recognition errors by 17.5 percent while running in real time on a single CPU core.
desk verdict A solid engineering paper whose stage-1 result holds up; the stage-2 real-IR finetuning claim needs a disclosure about whether the IRs and test recordings share the same cabin. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing mechanism is the dual-mask streaming MVDR estimator. Speech and noise masks, estimated by a causal network, build the target and interference spatial covariance matrices; the MVDR weight vector then filters each zone's microphone mixture with a distortionless constraint, so the output preserves the target speaker's spectral shape instead of carrying the nonlinear artifacts of direct neural separation. Around this core, the network uses three encoders (spectrogram, log power spectrum, and interaural phase difference between the two front microphones), full-band LSTM plus a time-skipped transform-average-concatenate (TAC) channel-fusion module, and a sub-band conformer. Traini
What would settle it
A cross-cabin test: finetune CabinSep-L with real-recorded impulse responses from one vehicle, then evaluate on real recordings from a second vehicle with a different cabin layout and microphone positions. If the 17.5% CER reduction and the non-standard-posture positioning jump from 60.4% to 98.9% collapse toward the simulated-IR-only numbers, the stage-2 advantage is largely environment adaptation rather than a general method.
Extended reading notes
Core claim
The central claim is that a mask-based MVDR speech separator can be made light enough for real-time in-car use and accurate enough to improve ASR on real recordings. CabinSep estimates one speech mask and one noise mask per zone, forms spatial covariance matrices from them, and applies the distortionless MVDR filter at inference instead of directly using the network output as the separated signal. With 0.4 GMACs and 0.21 RTF, the smallest variant CabinSep-S reduces average character error rate by 17.5% relative to DualSep-L when scored by WeNet, and by 14.2% when scored by SenseVoice; larger variants improve further. Adding real-recorded impulse responses in a 'mixed' augmentation—real IRs f
Load-bearing premise
The paper's strongest numbers combine real-recorded impulse responses in training with real-recorded test audio, and it never says whether the impulse responses and test recordings share the same car, microphone mounts, or recording session; if they do, part of the reported gain could be adaptation to that one cabin rather than generalizable improvement.
Editorial extensions
If this is right
- Because the system is causal and runs at 0.4 GMACs with a 0.21 real-time factor on a single car CPU, it can be deployed as a plug-and-play front end before an existing ASR model.
- Using MVDR at inference rather than the raw network output keeps separated speech ASR-friendly, as shown by consistent CER gains across two different frozen ASR backends.
- The time-skip operation halves TAC complexity with only a 0.41% average CER increase, making channel-aware separation affordable on constrained hardware.
- Mixed real/simulated impulse-response augmentation specifically fixes the boundary-speaker failure mode, lifting non-standard-posture zone positioning accuracy from 60.4% to above 90%.
- Larger CabinSep variants trade compute for accuracy, so the same architecture can scale with the available hardware budget.
Reading between the lines
- The stage-2 gains may be partly environment adaptation: if the real-recorded IRs and the real test recordings came from the same car and microphone mounts, the reported 17.5% CER gain and NSPA jump could shrink on a different cabin. A cross-cabin evaluation would settle this.
- Because interaural phase difference is used only between the two front microphones, rear-zone separation relies more on level and spectral cues; adding rear-microphone phase features could yield further gains for back-seat speech.
- The 'mixed real-recorded IRs' strategy suggests that the target zone's own early reflections matter most for zone positioning. If true, a lightweight calibration from a few in-cabin recordings could replace a full IR measurement campaign.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes CabinSep, a causal low-latency multi-channel speech separation front-end for in-car ASR. The architecture combines spec/LPS/IPD encoders, stacked 'full-sub' modules (full-band LSTM, a time-skip TAC, and sub-band conformer), dual speech/noise mask estimation, and a streaming mask-based MVDR at inference. Training is two-stage: stage 1 uses simulated image-source IRs; stage 2 finetunes with a 'mixed real-recorded IRs' augmentation in which the speaker's own zone channel uses a real measured IR and the other channels use simulated IRs. Evaluation is on real-recorded in-car audio from an electric vehicle, using WeNet and SenseVoice ASR CER plus a 'non-standard posture' positioning accuracy (NSPA). The headline results are that CabinSep-S (0.4 GMACs, 0.21 RTF on an automotive CPU) yields 17.5% and 14.2% relative CER reductions over DualSep-L for WeNet and SenseVoice, and that stage-2 finetuning raises NSPA from 60.4% to up to 98.9% while giving a small CER reduction.
Significance. If the claims hold, CabinSep is a practically valuable low-compute in-car separator: it improves ASR over a strong SOTA baseline at substantially lower cost and also addresses zone-level speaker positioning. The paper has genuine strengths: a real-world test set, two independent ASR back-ends, clear component ablations, and concrete efficiency numbers (GMACs and RTF on an automotive CPU). The main novelty lies in the mixed real/simulated IR augmentation strategy and the time-skip TAC complexity reduction. The most important risk is that the stage-2 generalization claim may be overstated because the relationship between the real-recorded IRs and the real-recorded test set is never disclosed. In addition, the headline NSPA metric is never defined, and all results are single-run point estimates, leaving the smaller ablative differences unquantified.
major comments (4)
- [§4.4, Table 2] NSPA is never defined. The paper only glosses it as 'positioning accuracy rate in non-standard posture' and reports percentages, but there is no formula, no description of how a zone decision is produced from the separator outputs (e.g., per-utterance energy, mask-based classification), and no labeling criterion. Since the stage-2 claim (60.4% to 98.9%) is a central advertised contribution, this metric must be specified precisely; otherwise the numbers are not reproducible even if data were available.
- [§4.1, §4.2, Table 2] The relationship between the real-recorded IRs used for stage-2 finetuning and the real-recorded test set is undisclosed. Section 4.1 reports 156 real IRs measured in car seats and a separate real-recorded EV test set (7.4 h + 4.9 h), but never states whether these share the same cabin, microphone positions, or recording session. If they do, stage-2 finetuning has directly observed the test environment's transfer functions, so the NSPA jump and the CER reduction in Table 2 would reflect adaptation to that cabin rather than evidence that the augmentation method generalizes. The manuscript must state whether IRs and test recordings are from the same or different cabins/sessions and, ideally, evaluate stage 2 on a held-out cabin or a matched-simulated condition.
- [Tables 1 and 2] All reported CER and NSPA numbers are single-point estimates with no confidence intervals, multiple seeds, or significance tests. The large headline gaps (e.g., CabinSep-S vs DualSep-L) are presumably robust, but several claims rely on small differences: the 0.41% CER increase with time-skip (7-2 vs 7-1), the 0.09% increase from chunking (7-7), and the 0.1-0.2% differences among IR augmentation variants in Table 2. These are within typical run-to-run or content-sampling variability. Please provide multiple trials or utterance-level paired significance tests (e.g., bootstrap or McNemar) for the main comparisons and ablations.
- [§4.3, Table 1] The baseline comparison may not be entirely fair. DualSep-S and DualSep-L are retrained on the same data, but no tuning protocol is reported (learning-rate schedule, epochs, early stopping, hyperparameter search). Worse, DualSep-L is altered by replacing its non-causal IVA with a causal IVA, and the impact of that substitution is not measured. A baseline with suboptimally tuned hyperparameters or a non-native causal variant could understate DualSep's performance. Please report the baseline tuning procedure and, if possible, include the original non-causal DualSep-L as an upper-bound reference.
minor comments (5)
- [§3.3] The claim that TAC is 'insensitive to time frames, so dropping every other frame ... is nearly lossless' is stated as fact, but it is a design assumption; the ablation (7-2) actually shows a 0.41% CER increase. Please soften the wording and explicitly tie it to the ablation result.
- [Table 2] The row labels ESS/MLS/TSP are not explained in the caption or in the table itself; state that these are the three activation signals used to measure real IRs.
- [§3.4] The loss weights are given as α=0.01, β=1, γ=0.01, but the text says 'to balance the magnitude' without justifying the chosen values or reporting sensitivity. A sentence on how these were selected would help.
- [Throughout] Typographical issues: 'time-streched pulses' (§4.1), 'recieved' and 'micriphone' (§2), 'Refering' (§3.3), 'conformerr' (§3.3), 'to a great extend' (§1), and reference [30] begins with 'Fneural' instead of 'FullNeural'.
- [Table 1, row 7-7] The text describes 7-7 as 'adding chunks' and then says it limits the conformer to look back at a maximum of 2 seconds. This is confusing: 'chunk' usually refers to input segmentation, while the described operation is a memory/look-back constraint. Clarify what is being added.
Circularity Check
No significant circularity: central claims are empirical comparisons and standard architecture components; the few self-citations are peripheral.
full rationale
The paper's central claims are measured CER values and computational-cost comparisons (Tables 1 and 2), not derivations from fitted constants. The CabinSep architecture uses standard STFT/LPS/IPD features, mask estimation trained with the supervised loss in Eq. (7), and MVDR beamforming in Eqs. (3)-(4); none of these equations is defined in terms of the reported CER or NSPA outcomes, so there is no self-definitional reduction. The headline 17.5% relative CER reduction is a head-to-head empirical comparison with DualSep-L on a real-recorded test set, with baselines trained on the same data; it is not a fitted parameter renamed as a prediction. The stage-2 real-recorded-IR finetuning study (Table 2) is likewise an empirical augmentation comparison. A data-coupling concern exists—the paper never states whether the real-recorded IRs and the real-vehicle test set share the same cabin, microphone positions, or recording session—but this is a soundness/data-hygiene issue rather than a definitional circularity, and no passage asserts the relevant identity, so no quote-level reduction can be exhibited. The same-group citations ([8], [23], [35]) are used as background or general references for ASR distortion and conformer blocks; they are not load-bearing, and no uniqueness theorem or ansatz is imported from the authors' prior work. NSPA is undefined, but that is a reporting/checkability gap, not a circular derivation. Therefore no circular step is identified.
Assumptions & free parameters
free parameters (7)
- loss weights alpha, beta, gamma =
0.01, 1, 0.01
- full-sub module count N =
1, 2, 3 for S/M/L
- TAC channel compression ratio d =
4 for S/M, 2 for L
- conformer layer count =
4 in S, 2 in M and L
- noise augmentation SNR ranges =
background [-20,25] dB, transient [-5,5] dB
- streaming chunk look-back =
2 seconds in ablation 7-7
- learning rate schedule =
1e-4, halved every 20k steps
assumptions (4)
- domain assumption Each zone contains at most one active speaker, and the multi-channel mixture is a linear convolution of zone-clean sources with impulse responses plus noise (Eq. 1-2).
- domain assumption Interaural phase difference is usable only for the front microphone pair because back-row spacing causes spatial aliasing (Sec. 3.2).
- ad hoc to paper TAC is insensitive to time-frame order, so dropping every other frame before TAC and recombining is nearly lossless (Sec. 3.3).
- domain assumption Real-recorded impulse responses measured in car seats are representative of the real-recorded test conditions (Sec. 4.1-4.2).
Cite this review
Pith. "Pith review of CabinSep: IR-Augmented Mask-Based MVDR for Real-Time In-Car Speech Separation with Distributed Heterogeneous Arrays." pith.science (2026). https://pith.science/paper/HFIYTYIB
@misc{pith2026250901399,
author = {Pith},
title = {Pith review of: CabinSep: IR-Augmented Mask-Based MVDR for Real-Time In-Car Speech Separation with Distributed Heterogeneous Arrays},
year = {2026},
howpublished = {\url{https://pith.science/paper/HFIYTYIB}},
note = {Machine review of arXiv:2509.01399}
}
read the original abstract
Separating overlapping speech from multiple speakers is crucial for effective human-vehicle interaction. This paper proposes CabinSep, a lightweight neural mask-based minimum variance distortionless response (MVDR) speech separation approach, to reduce speech recognition errors in back-end automatic speech recognition (ASR) models. Our contributions are threefold: First, we utilize channel information to extract spatial features, which improves the estimation of speech and noise masks. Second, we employ MVDR during inference, reducing speech distortion to make it more ASR-friendly. Third, we introduce a data augmentation method combining simulated and real-recorded impulse responses (IRs), improving speaker localization at zone boundaries and further reducing speech recognition errors. With a computational complexity of only 0.4 GMACs, CabinSep achieves a 17.5% relative reduction in speech recognition error rate in a real-recorded dataset compared to the state-of-the-art DualSep model. Demos are available at: https://cabinsep.github.io/cabinsep/.
Figures
Reference graph
Works this paper leans on
-
[1]
Introduction Speech interaction is crucial for in-car intelligence, with auto- matic speech recognition (ASR) as a key gateway for human- vehicle interaction. Speech recognition accuracy directly im- pacts interaction efficiency and user experience [1, 2, 3, 4, 5]. However, when multiple passengers interact with the car simul- taneously, overlapping speec...
work page Pith review arXiv 2025
-
[2]
Problem Formulation As shown in Figure 1(a), we focus on in-car speech separation. We divide the car cabin into Z zones, with each zone corre- sponding to a single channel microphone, and at most one per- son speaks in each zone. Suppose there are P person speaking in the car ( P ≤ Z), the clean speech corresponding to each zone is set as s(z), and the si...
-
[3]
Method 3.1. System overview The overall architecture of our proposed CabinSep is shown in Figure 1(b). First, the Z-channel audio mixture y is trans- formed by the Short-Time Fourier Transform (STFT) to obtain the T-F spectrum Y ∈ CZ×T ×F , where Z represents the chan- nel dimension, T represents the time dimension, and F repre- sents the frequency dimens...
-
[4]
Experiment 4.1. Datasets Training set : The training set includes clean speech, back- ground noise, transient noises including claps, coughs, etc., and IRs used to simulate reverberation. The clean speech is from AISHELL-2 [37] and augmented to simulate scenarios with one to four passengers speaking simultaneously in a car. The background and transient no...
-
[5]
Conclusions This paper proposes an in-car speech separation approach with excellent generalization ability, enhancing the back-end ASR models’ performance. We validate the system using two dif- ferent ASR models, WeNet and SenseV oice, without any joint training, demonstrating its plug-and-play capability and com- patibility with various ASR systems. Trai...
-
[6]
M. Capallera, L. Angelini, and et al., “Human-vehicle interac- tion to support driver’s situation awareness in automated vehicles: A systematic review,” IEEE Trans. Intell. Veh., vol. 8, no. 3, pp. 2551–2567, 2023
work page 2023
-
[7]
Intelligent in-vehicle interac- tion technologies,
P. K. Murali, M. Kaboli, and et al., “Intelligent in-vehicle interac- tion technologies,” Adv. Intell. Syst., vol. 4, no. 2, 2022
work page 2022
-
[8]
Model-based estimation of in- car-communication feedback applied to speech zone detection,
K. M ¨uller, S. Doclo, and et al., “Model-based estimation of in- car-communication feedback applied to speech zone detection,” in IWAENC. IEEE, 2022, pp. 1–5
work page 2022
Show all 45 references
-
[9]
Conversational in-vehicle dialog systems: The past, present, and future,
F. Weng, P. Angkititrakul, and et al., “Conversational in-vehicle dialog systems: The past, present, and future,” IEEE Signal Pro- cess. Mag., vol. 33, no. 6, pp. 49–60, 2016
2016
-
[10]
Adaptive nonlinear regression using multiple distributed microphones for in-car speech recog- nition,
W. Li, C. Miyajima, and et al., “Adaptive nonlinear regression using multiple distributed microphones for in-car speech recog- nition,” IEICE Trans. Fundam. Electron. Commun. Comput. Sci., vol. 88-A, no. 7, pp. 1716–1723, 2005
2005
-
[11]
The third ’chime’ speech sepa- ration and recognition challenge: Dataset, task and baselines,
J. Barker, R. Marxer, and et al., “The third ’chime’ speech sepa- ration and recognition challenge: Dataset, task and baselines,” in ASRU. IEEE, 2015, pp. 504–511
2015
-
[12]
Audio-visual multi-channel integration and recognition of overlapped speech,
J. Yu and S. Zhang, “Audio-visual multi-channel integration and recognition of overlapped speech,” IEEE ACM Trans. Audio Speech Lang. Process., vol. 29, pp. 2067–2082, 2021
-
[13]
The NPU-ASLP system for audio- visual speech recognition in MISP 2022 challenge,
P. Guo, H. Wang, and et al., “The NPU-ASLP system for audio- visual speech recognition in MISP 2022 challenge,” in ICASSP. IEEE, 2023, pp. 1–2
2022
-
[14]
Automatic channel selection and spatial feature integration for multi-channel speech recognition across various array topologies,
B. Mu, P. Guo, and et al., “Automatic channel selection and spatial feature integration for multi-channel speech recognition across various array topologies,” in ICASSP. IEEE, 2024, pp. 11 396– 11 400
2024
-
[15]
Neural blind source separa- tion and diarization for distant speech recognition,
Y . Bando, T. Nakamura, and et al., “Neural blind source separa- tion and diarization for distant speech recognition,”arXiv preprint arXiv:2406.08396, 2024
2024 arXiv
-
[16]
Recognizing overlapped speech in meetings: A multichannel separation approach using neural networks,
T. Yoshioka, H. Erdogan, and et al., “Recognizing overlapped speech in meetings: A multichannel separation approach using neural networks,” in Interspeech, B. Yegnanarayana, Ed. ISCA, 2018, pp. 3038–3042
2018
-
[17]
Neural network based spectral mask estimation for acoustic beamforming,
J. Heymann, L. Drude, and et al., “Neural network based spectral mask estimation for acoustic beamforming,” in ICASSP. IEEE, 2016, pp. 196–200
2016
-
[18]
Improved MVDR beam- forming using single-channel mask prediction networks,
H. Erdogan, J. R. Hershey, and et al., “Improved MVDR beam- forming using single-channel mask prediction networks,” in In- terspeech, N. Morgan, Ed. ISCA, 2016, pp. 1981–1985
2016
-
[19]
Joint training of complex ratio mask based beamformer and acoustic model for noise robust asr,
Y . Xu, C. Weng, and et al., “Joint training of complex ratio mask based beamformer and acoustic model for noise robust asr,” in ICASSP. IEEE, 2019, pp. 6745–6749
2019
-
[20]
A fast-converging adaptive frequency- domain MVDR beamformer for speech enhancement,
S. Zhao and D. L. Jones, “A fast-converging adaptive frequency- domain MVDR beamformer for speech enhancement,” inINTER- SPEECH. ISCA, 2012, pp. 1930–1933
2012
-
[21]
Multi-channel multi-frame ADL- MVDR for target speech separation,
Z. Zhang, Y . Xu, and et al., “Multi-channel multi-frame ADL- MVDR for target speech separation,” IEEE ACM Trans. Audio Speech Lang. Process., vol. 29, pp. 3526–3540, 2021
2021
-
[22]
Multi-microphone complex spec- tral mapping for utterance-wise and continuous speech separa- tion,
Z. Wang, P. Wang, and et al., “Multi-microphone complex spec- tral mapping for utterance-wise and continuous speech separa- tion,” IEEE ACM Trans. Audio Speech Lang. Process. , vol. 29, pp. 2001–2014, 2021
2001
-
[23]
Embedding and beamforming: All- neural causal beamformer for multichannel speech enhancement,
A. Li, W. Liu, and et al., “Embedding and beamforming: All- neural causal beamformer for multichannel speech enhancement,” in ICASSP. IEEE, 2022, pp. 6487–6491
2022
-
[24]
End-to-end microphone permuta- tion and number invariant multi-channel speech separation,
Y . Luo, Z. Chen, and et al., “End-to-end microphone permuta- tion and number invariant multi-channel speech separation,” in ICASSP. IEEE, 2020, pp. 6394–6398
2020
-
[25]
Dualsep: A light-weight dual-encoder convolutional recurrent network for real-time in-car speech sepa- ration,
Z. Wang, J. Sun, and et al., “Dualsep: A light-weight dual-encoder convolutional recurrent network for real-time in-car speech sepa- ration,” in SLT. IEEE, 2024, pp. 286–293
2024
-
[26]
Zoneformer: On-device neu- ral beamformer for in-car multi-zone speech separation, enhance- ment and echo cancellation,
Y . Xu, V . Kothapally, and et al., “Zoneformer: On-device neu- ral beamformer for in-car multi-zone speech separation, enhance- ment and echo cancellation,” in Interspeech. ISCA, 2023, pp. 5117–5121
2023
-
[27]
SDR - half-baked or well done?
J. L. Roux, S. Wisdom, and et al., “SDR - half-baked or well done?” in ICASSP. IEEE, 2019, pp. 626–630
2019
-
[28]
An audio-quality-based multi-strategy approach for target speaker extraction in the misp 2023 chal- lenge,
R. Han, X. Yan, and et al., “An audio-quality-based multi-strategy approach for target speaker extraction in the misp 2023 chal- lenge,” in ICASSP-Workshops. IEEE, 2024, pp. 27–28
2023
-
[29]
Single channel tar- get speaker extraction and recognition with speaker beam,
M. Delcroix, K. Zmol ´ıkov´a, and et al., “Single channel tar- get speaker extraction and recognition with speaker beam,” in ICASSP. IEEE, 2018, pp. 5554–5558
2018
-
[30]
Impulse response data augmentation and deep neu- ral networks for blind room acoustic parameter estimation,
N. J. Bryan, “Impulse response data augmentation and deep neu- ral networks for blind room acoustic parameter estimation,” in ICASSP. IEEE, 2020, pp. 1–5
2020
-
[31]
Wenet: Production oriented stream- ing and non-streaming end-to-end speech recognition toolkit,
Z. Yao, D. Wu, and et al., “Wenet: Production oriented stream- ing and non-streaming end-to-end speech recognition toolkit,” in Interspeech. ISCA, 2021, pp. 4054–4058
2021
-
[32]
Funaudiollm: V oice understanding and generation foundation models for natural interaction between humans and llms,
K. An, Q. Chen, and et al., “Funaudiollm: V oice understanding and generation foundation models for natural interaction between humans and llms,” arXiv preprint arXiv:2407.04051, 2024
2024 arXiv
-
[33]
ICMC-ASR: the ICASSP 2024 in-car multi-channel automatic speech recognition challenge,
H. Wang, P. Guoand, and et al., “ICMC-ASR: the ICASSP 2024 in-car multi-channel automatic speech recognition challenge,” in ICASSP-Workshops. IEEE, 2024, pp. 63–64
2024
-
[34]
Multi-channel deep cluster- ing: Discriminative spectral and spatial embeddings for speaker- independent speech separation,
Z. Wang, J. L. Roux, and et al., “Multi-channel deep cluster- ing: Discriminative spectral and spatial embeddings for speaker- independent speech separation,” in ICASSP. IEEE, 2018, pp. 1–5
2018
-
[35]
Fneural speech enhancement with very low algorithmic latency and complexity via integrated full-and sub-band modeling,
Z.-Q. Wang, S. Cornell, and et al., “Fneural speech enhancement with very low algorithmic latency and complexity via integrated full-and sub-band modeling,” in ICASSP. IEEE, 2023, pp. 1–5
2023
-
[36]
Frame-by-frame closed-form update for mask-based adaptive mvdr beamforming,
T. Higuchi, K. Kinoshita, and et al., “Frame-by-frame closed-form update for mask-based adaptive mvdr beamforming,” in ICASSP, 2018, pp. 531–535
2018
-
[37]
Conformer: Convolution-augmented transformer for speech recognition,
A. Gulati, J. Qin, and et al., “Conformer: Convolution-augmented transformer for speech recognition,” inInterspeech. ISCA, 2020, pp. 5036–5040
2020
-
[38]
Continuous speech separation with conformer,
S. Chen, Y . Wu, and et al., “Continuous speech separation with conformer,” in ICASSP. IEEE, 2021, pp. 5749–5753
2021
-
[39]
Uformer: A unet based dilated complex & real dual-path conformer network for simultaneous speech en- hancement and dereverberation,
Y . Fu, Y . Liu, and et al., “Uformer: A unet based dilated complex & real dual-path conformer network for simultaneous speech en- hancement and dereverberation,” in ICASSP. IEEE, 2022, pp. 7417–7421
2022
-
[40]
Distil-dccrn: A small-footprint DC- CRN leveraging feature-based knowledge distillation in speech enhancement,
R. Han, W. Xu, and et al., “Distil-dccrn: A small-footprint DC- CRN leveraging feature-based knowledge distillation in speech enhancement,” IEEE Signal Process. Lett. , vol. 31, pp. 2075– 2079
-
[41]
Multirate systems and filter banks,
T. Saramaki and R. Bregovi ´c, “Multirate systems and filter banks,” 2002. [Online]. Available: https://api.semanticscholar. org/CorpusID:9768753
2002
-
[42]
Aishell-2: Transforming mandarin asr research into industrial scale,
J. Du, X. Na, and et al., “Aishell-2: Transforming mandarin asr research into industrial scale,” arXiv preprint arXiv:1808.10583, 2018
2018 arXiv
-
[43]
Simultaneous measurement of impulse response and distortion with a swept-sine technique,
A. Farina, “Simultaneous measurement of impulse response and distortion with a swept-sine technique,” in Audio engineering so- ciety convention 108. Audio Engineering Society, 2000
2000
-
[44]
A fast hadamard transform method for the evalua- tion of measurements using pseudrandom test signals,
H. Alrutz, “A fast hadamard transform method for the evalua- tion of measurements using pseudrandom test signals,”Proc. 11th ICA, Paris, July 1983, vol. 6, pp. 235–238, 1983
1983
-
[45]
Computer-generated pulse signal applied for sound measurement,
N. Aoshima, “Computer-generated pulse signal applied for sound measurement,” The Journal of the Acoustical Society of America, vol. 69, no. 5, pp. 1484–1488, 1981
1981
Reviewed August 5, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.