Pith. sign in

REVIEW 3 major objections 4 minor 29 references

Improving endpoint detection in end-to-end streaming ASR for conversational speech

T0 review · 3 major / 4 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 The paper claims that an end-of-word token, trained with a delay penalty and synchronized with a voice-activity detector, lets a streaming transducer ASR endpoint conversational speech at its no-endpointing oracle word error rate.

desk verdict A useful engineering increment whose headline WER claim is not actually supported by the reported data — the 'oracle' comparison is missing and the EOW rule's latency is uncontrolled. read the letter →

arxiv 2505.17070 v1 pith:JHWBA6GR submitted 2025-05-19 cs.CL cs.AIcs.SDeess.AS

Anandh C , Karthik Pandia Durai , Jeena Prakash , Manickavela Arumugam , Kadri Hacioglu , S.Pavankumar Dubagunta , Andreas Stolcke , Shankar Venkatesan
show 1 more author
Aravind Ganapathiraju
This is my paper · ORCID
classification cs.CLcs.AIcs.SDeess.AS
keywords endpointingstreamingASRRNN-Tvoiceactivitydetectionend-of-wordtokenlatencySwitchboardZipformer
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 tries to solve a user-facing problem in streaming speech recognition: deciding exactly when a speaker has finished, without cutting off the last word and without waiting too long. For transducer-based models, endpointing usually relies on runs of blank tokens, but delayed token emission makes blank runs unreliable. The paper claims that two modifications fix this: an auxiliary voice-activity detection network that reads the ASR encoder's internal representation, and an end-of-word token added to every word during training with a delay penalty. On conversational telephone speech, the end-of-word endpoint rule reaches a word error rate of 21.4%, the same as the model's no-endpointing oracle (the lower bound when no endpoint is ever triggered), while endpoint precision rises from 47.7% to 81.4% compared with blank-based endpointing. A reader should care because endpointing errors are what make voice agents interrupt users or return truncated transcripts, and these results suggest those errors can be cut sharply at low latency.

What carries the argument

The central machinery is a pair of synchronized signals for the endpoint decision. The first is encNET, a two-hidden-layer feed-forward network that takes the Zipformer encoder's frame embeddings and emits a speech/non-speech label; the duration of contiguous non-speech frames from this network triggers a candidate endpoint. The second is an end-of-word (EOW) token appended to every word in the training transcripts, trained with a delay penalty so the transducer emits EOW promptly at word endings. The endpoint is finalized only when the most recent output from the transducer's joiner is EOW, which prevents the VAD's early silence detection from cutting a word in half while the delayed token is still on its way. The delay penalty removes the blank-token lag that causes false, missed, and delayed endpoints in the baseline.

What would settle it

A single ablation study would settle it: train four variants of the system--encNET trained with teacher labels versus hand-corrected labels, and with versus without the EOW token--and compare endpoint precision and WER. If the gains disappear when teacher labels are replaced by hand-corrected labels, the VAD teacher is the source; if they disappear when EOW is removed, the token is the source.

Watch

Extended reading notes

Core claim

The central discovery, as the paper states it, is that an end-of-word token is the load-bearing element for high-precision endpointing in a delay-penalized transducer. The EOW token is appended to each word in the training text, so the transducer's output layer must emit a special symbol at the true word boundary. When the endpoint logic waits for that token instead of counting blank frames, false endpoints caused by delayed non-blank emission disappear, and WER on the Switchboard test set improves to 21.4%--exactly the oracle WER of the same model with no endpointing. Adding the VAD condition (trailing silence) alongside EOW gives the best F1 score of 73.0 at a small WER cost (21.7%). The paper also reports that a VAD trained on encoder embeddings (encNET) has an equal error rate of 0.105 versus 0.182 for a VAD trained on acoustic features, showing that the ASR encoder carries reliable speech/non-speech information.

Load-bearing premise

The approach relies on the external VAD teacher's speech/non-speech labels being a trustworthy reference for conversational audio; if those labels are biased, encNET inherits the bias and the reported endpoint gains may reflect the teacher rather than the ASR encoder embeddings.

Editorial extensions

If this is right

  • With EOW gating, a 200 ms trailing-silence threshold no longer forces a WER penalty: EOW-based endpointing reaches oracle WER at that aggressive setting, whereas blank-based and TS-only endpointing degrade.
  • The VAD branch can be trained from the ASR encoder's own embeddings, so endpoint detection comes at little extra compute and shares the ASR model's acoustic context.
  • Combining TS and EOW rules gives production flexibility: TS+EOW produces the best F1 (73.0) at slight WER cost (21.7%), useful when avoiding missed endpoints matters more than WER.
  • The latency-WER curve saturates around 600 ms of trailing silence for all systems, so EOW shifts the usable operating point to much lower latency without transcription loss.
  • Reaching oracle WER under endpointing means endpoint decisions can be made without sacrificing ASR accuracy on spontaneous conversational speech.

Reading between the lines

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

  • A natural next test is to drop the external teacher entirely and train encNET from the ASR model's own blank/non-blank emission history; if that self-supervised VAD matches teacher-trained encNET, the endpointing pipeline becomes fully self-contained.
  • Because EOW forces a lexical boundary at word end, the same token could serve as a cheap word-boundary signal for word-level timestamps, segmentation, or turn-taking prediction in streaming systems, not just endpointing.
  • The precision/recall trade-off suggests an assistant-oriented deployment would use EOW for the primary endpoint and accept lower recall, recovering missed endpoints with a downstream backchannel or turn-taking model; the paper does not test this combination.
  • Languages with long or morphologically complex words may not align acoustic word boundaries with the EOW training token as cleanly as English, so the 21.4% result on Switchboard should be re-measured on such languages before generalizing.
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

3 major / 4 minor

Summary. The paper addresses endpointing in streaming transducer-based ASR for conversational speech. It proposes two components: an auxiliary VAD network that uses either mel features (melNET) or encoder embeddings (encNET) to trigger endpoints from trailing silence, and an end-of-word (EOW) token trained with a delay penalty to promote prompt emission of word-final tokens. Three endpointing strategies are compared: trailing-silence only (TS), EOW-only, and TS+EOW. Experiments on Switchboard (train/validation/test: 2353/132/19 calls, with 3.1 h of test audio) report endpoint precision/recall/F1 and WER at 200 ms trailing silence, plus latency-WER curves for 400-800 ms. The main reported results are that encNET gives EER 0.105 versus 0.182 for melNET, that encNET TS endpointing raises F1 from 46.6 to 78.6 and lowers WER from 24.4% to 23.0%, and that the EOW rule achieves 21.4% WER, which the text calls 'oracle WER for this model'.

Significance. If the claims hold, the paper gives a practical and inexpensive way to improve endpointing in streaming conversational ASR: reusing encoder embeddings for a lightweight VAD branch, and adding an EOW token with a delay penalty. Strengths of the study are that it evaluates on spontaneous Switchboard speech rather than read speech, performs call-level decoding, uses the sed_eval toolkit with a 200 ms tolerance, and includes a blank-based baseline and an oracle for comparison. The DET-curve comparison between encNET and melNET is informative. However, the central EOW result is under-supported: the oracle WER for the EOW-trained model is not reported, the 200 ms operating point is missing from the latency plot, and the small test set has no significance tests. These issues are fixable but currently weaken the main conclusion.

major comments (3)
  1. [Section 3.4 / Table 3] The statement that EOW-based endpointing 'achiev[es] oracle WER for this model (21.4%)' cannot be verified from the reported data. Table 2 gives the no-endpointing oracle WER for the vanilla model as 21.92%, but no oracle WER for the EOW-trained model is reported anywhere, and Table 3 has no 'no endpointing' row for that model. The authors should either measure and report the EOW model's oracle WER explicitly, or soften the claim to a direct comparison with the other endpointing strategies.
  2. [Section 3.4 / Figure 5 and Section 2.3] The WER comparison for the EOW rule at the 200 ms trailing-silence setting is confounded by latency. As described in Section 2.3 for cases C1 and C2, an endpoint under the EOW rule is finalized only when the joiner emits EOW; if EOW emission occurs after the 200 ms silence threshold, the actual endpoint latency is longer than 200 ms. Figure 5 plots WER only for 400, 600, and 800 ms and shows all systems approaching the oracle by 600 ms, so the 21.4% WER could reflect a longer effective latency rather than better endpointing. The authors should report the actual endpoint-latency distribution for each strategy at the 200 ms setting and compare WER at matched latency.
  3. [Section 3.1 / Tables 1-3] The statistical basis is thin: the test set contains only 19 calls (3.1 h), and no confidence intervals or significance tests are reported for the precision/recall/F1/WER differences. Because several of the headline differences are small (e.g., 21.4% vs. 21.7%, or 23.0% vs. 21.92%), the authors should add bootstrap confidence intervals or significance tests and explicitly state that the delay-penalty coefficient, the EER operating point, and the trailing-silence thresholds were selected on the validation set only.
minor comments (4)
  1. [Section 3.3] The text says 'improving the WER (from 22.4% to 22.9%)', but Table 2 shows the blank-based vanilla baseline WER is 24.4%; this appears to be a typo and should be corrected to 24.4%.
  2. [Section 3.3] The phrase 'with out endpointing' should read 'without endpointing'.
  3. [Section 3.1] The definition of the reference endpoint events used in the sed_eval alignment is not stated; the authors should specify how reference timestamps were derived from the call-level transcripts, since precision/recall/F1 depend directly on this definition.
  4. [Figure 5] The figure omits the 200 ms operating point that is central to the discussion in Sections 3.3 and 3.4; adding this point (or explaining why it is not shown) would make the latency-WER trade-off much clearer.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's endpointing improvements are empirical comparisons against baselines and an oracle, with no fitted parameter or derivation chain that reduces to its own inputs.

full rationale

The paper is an empirical endpointing study, not a derivational one. Its central claims are that an auxiliary VAD network (encNET/melNET) trained on ASpIRE labels improves endpoint precision, and that an end-of-word token combined with a delay penalty improves the latency/WER trade-off. These are evaluated by computing precision, recall, F1, and WER against blank-based baselines and an oracle without endpointing (Table 2 and Table 3). No equation or construction in the paper defines the reported endpointing gains in terms of the method's own inputs; the VAD operating point is chosen by EER on a validation set, and the EOW rule is a discrete endpointing policy, not a fitted quantity renamed as a prediction. The test set is taken from Stolcke et al. [23], which includes one of the paper's co-authors, but that citation is used only as a source of an external Switchboard test partition and does not supply the endpointing result. The paper's claim that EOW endpointing 'achieves oracle WER for this model (21.4%)' is not fully checkable from Table 3 because no no-endpointing row is given for the EOW-trained model, and Figure 5 only plots latency settings of 400, 600, and 800 ms; this is a reporting or control-comparison gap, not circularity. There is no self-definitional step, no fitted input presented as a prediction, and no load-bearing self-citation chain. Therefore the circularity score is 0.

Assumptions & free parameters 4 free parameters · 4 assumptions · 1 invented entities

The central claim rests on standard ASR model behavior, a teacher VAD's label quality, and two tuned thresholds (delay penalty and trailing silence). The EOW token is a modeling construct introduced here; no external falsifiable evidence is provided for it beyond the reported system comparisons.

free parameters (4)
  • delay penalty coefficient = 0.0004
    Selected by WER criterion in Section 3.3; the value is tuned on data and affects emission timing and EOW promptness.
  • VAD operating threshold = EER operating point (encNET EER 0.105)
    Chosen on the validation set via the DET curve in Section 3.2; changes the endpoint trigger rate.
  • trailing silence duration = 200 ms primary; 400/600/800 ms swept
    Main results use 200 ms, and Figure 5 sweeps other values to show the latency-WER trade-off.
  • baseline blank-frame threshold = 6 blank frames
    Figure 1 specifies that 6 consecutive blank frames trigger an endpoint in the baseline; this is a baseline hyperparameter.
assumptions (4)
  • domain assumption The Zipformer transducer and the delay penalty loss from [11] behave as described in the cited literature.
    The paper relies on Icefall and [11] without re-deriving or verifying these components; standard ASR modeling background.
  • domain assumption The Switchboard test set from [23] and the sed_eval toolkit with 200 ms tolerance provide trustworthy endpoint event references.
    Used for precision and recall; the evaluation method is standard in sound event detection but is applied here to endpointing.
  • domain assumption ASpIRE VAD labels are a suitable teacher for frame-level speech or nonspeech despite a DER of 0.426 against Switchboard references.
    Section 3.2 justifies this because Switchboard annotations ignore cross-talk; if the teacher is biased, encNET inherits the bias.
  • ad hoc to paper The EOW token can be inserted after every word in training transcripts without corrupting the transducer alignment or language model.
    The paper proposes this modeling change but does not analyze its effect on alignment; it is specific to this method.
invented entities (1)
  • End-of-word (EOW) token
    purpose: Added after each word in training so the decoder can signal a complete word boundary and the endpointing logic can wait for it before cutting off.
    It is a modeling construct with no external falsifiable handle outside this paper's WER and latency measurements; its value is only assessed through the reported system performance.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improving endpoint detection in end-to-end streaming ASR for conversational speech." pith.science (2026). https://pith.science/paper/JHWBA6GR

@misc{pith2026250517070,
  author       = {Pith},
  title        = {Pith review of: Improving endpoint detection in end-to-end streaming ASR for conversational speech},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JHWBA6GR}},
  note         = {Machine review of arXiv:2505.17070}
}
read the original abstract

ASR endpointing (EP) plays a major role in delivering a good user experience in products supporting human or artificial agents in human-human/machine conversations. Transducer-based ASR (T-ASR) is an end-to-end (E2E) ASR modelling technique preferred for streaming. A major limitation of T-ASR is delayed emission of ASR outputs, which could lead to errors or delays in EP. Inaccurate EP will cut the user off while speaking, returning incomplete transcript while delays in EP will increase the perceived latency, degrading the user experience. We propose methods to improve EP by addressing delayed emission along with EP mistakes. To address the delayed emission problem, we introduce an end-of-word token at the end of each word, along with a delay penalty. The EP delay is addressed by obtaining a reliable frame-level speech activity detection using an auxiliary network. We apply the proposed methods on Switchboard conversational speech corpus and evaluate it against a delay penalty method.

Figures

Figures reproduced from arXiv: 2505.17070 by the authors.

Figure 1
Figure 1. represents the emission of non-blank or blank, respec￾tively, for each frame. A total of 6 frames with blank emissions is required to trigger an endpoint. T1, T2, and T3 are the times￾tamps where the acoustics corresponding to non-blank emis￾sions are present. Endpoint E1 is a false endpoint due to the delayed emission of NB from T1. Similarly, delayed emission of non-blank from T2 leads to missing of expected endpo… view at source ↗
Figure 2
Figure 2. Block diagram of proposed endpointing method used to trigger endpoints. Output from the transcription encoder network (Zipformer encoder) is used as input representation for VAD. A feed-forward neural network is used as a classifier and the output is binary. An endpoint is detected right after a con￾tiguous silence of x ms. The decision from the VAD classifier is used to obtain the duration of silence. A limitation … view at source ↗
Figure 3
Figure 3. C1 and C2 are the possible endpointing cases where the joiner output is emitted before (t2) or after (t3) the VAD decision (t1 + δ), respectively. Here δ is the trailing silence (after speech ended) required to trigger an endpoint. t1 + δ and t3 are the endpoints of the proposed method; t1 + δ and t3 + δ are the endpoints of the baseline for the cases C1 and C2, respectively. an error in the last word. To address th… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Detection error trade-off curves for encNET and mel￾NET. The EER for encNET and melNET are 0.105 and 0.182, respectively. The encNET is a feed-forward neural network with two hid￾den layers, trained to classify speech or nonspeech at the frame level. The encoding from …
Figure 5
Figure 5. Figure 5: Latency vs WER for different systems for various trail￾ing silence configurations (400ms, 600ms, and 800ms) In the above results, endpointing was triggered for a trailing silence of 200 ms. A more thorough comparison is possible by plotting the WER of proposed methods …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 22 canonical work pages

  1. [16]

    Alignment restricted stream- ing recurrent neural network transducer,

    J. Mahadeokar, Y . Shangguan, D. Le, G. Keren, H. Su, T. Le, C.-F. Yeh, C. Fuegen, and M. L. Seltzer, “Alignment restricted stream- ing recurrent neural network transducer,” in 2021 IEEE Spoken Language Technology Workshop (SLT). IEEE, 2021, pp. 52–59

    work page 2021

  2. [17]

    Reducing Streaming ASR Model Delay with Self Alignment,

    J. Kim, H. Lu, A. Tripathi, Q. Zhang, and H. Sak, “Reducing Streaming ASR Model Delay with Self Alignment,” in Proceed- ings of Interspeech, 2021, pp. 3440–3444

    work page 2021

  3. [1]

    Introduction Speech endpointing, i.e., the detection of end-of-speech in a turn, is an important aspect of conversational speech recogni- tion. Two-party human-to-human/machine conversations in an industrial setting, such as a customer speaking to a call cen- ter agent, or a customer speaking to a voice bot, require precise endpointing such that the live ...

    work page Pith review arXiv 2025

  4. [2]

    occasional

    Proposed methods We now describe the problem of endpointing delay and the pro- posed methods to address it. 2.1. Endpointing delay in transducer-based ASR In ASR systems, the trailing silence (non-speech following a speech region) is generally used to detect endpoint, though con- textual information from the preceding speech region and con- versation can ...

    work page

  5. [3]

    Datasets, experiments, and results 3.1. Dataset description and modelling framework We use the Switchboard corpus [19] in our experiments since it contains realistic spontaneous, human-human speech, as could be expected from natural human-machine dialog, and unlike pre-planned or read speech. The test set from [23] was used. The remaining data from the co...

    work page

  6. [4]

    A separate speech detector network operates in parallel with the ASR decoder to determine speech/non-speech at the frame-level

    Conclusion We have proposed methods to improve endpointing in end-to- end transducer-based ASR systems. A separate speech detector network operates in parallel with the ASR decoder to determine speech/non-speech at the frame-level. We show that ASR en- coder embedding can be reused for V AD for better results. Sec- ond, we introduce an end-of-word token i...

    work page

  7. [5]

    Towards end-to-end speech recognition with recurrent neural networks,

    A. Graves and N. Jaitly, “Towards end-to-end speech recognition with recurrent neural networks,” in International conference on machine learning. PMLR, 2014, pp. 1764–1772

    work page 2014

  8. [6]

    Sequence transduction with recurrent neural net- works,

    A. Graves, “Sequence transduction with recurrent neural net- works,” arXiv preprint arXiv:1211.3711, 2012

    arXiv 2012

Show all 29 references
  1. [7]

    End-to-end continuous speech recognition using attention-based recurrent nn: First results,

    J. Chorowski, D. Bahdanau, K. Cho, and Y . Bengio, “End-to-end continuous speech recognition using attention-based recurrent nn: First results,” arXiv preprint arXiv:1412.1602, 2014

    2014 arXiv

  2. [8]

    Online and linear-time attention by enforcing monotonic alignments,

    C. Raffel, M.-T. Luong, P. J. Liu, R. J. Weiss, and D. Eck, “Online and linear-time attention by enforcing monotonic alignments,” in International conference on machine learning. PMLR, 2017, pp. 2837–2846

    2017

  3. [9]

    Joint ctc-attention based end-to-end speech recognition using multi-task learning,

    S. Kim, T. Hori, and S. Watanabe, “Joint ctc-attention based end-to-end speech recognition using multi-task learning,” in2017 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, 2017, pp. 4835–4839

    2017

  4. [10]

    Conformer: Convolution-augmented Transformer for Speech Recognition,

    A. Gulati, J. Qin, C.-C. Chiu, N. Parmar, Y . Zhang, J. Yu, W. Han, S. Wang, Z. Zhang, Y . Wu, and R. Pang, “Conformer: Convolution-augmented Transformer for Speech Recognition,” in Proc. Interspeech 2020, 2020, pp. 5036–5040

    2020

  5. [11]

    Zipformer: A faster and better encoder for automatic speech recognition,

    Z. Yao, L. Guo, X. Yang, W. Kang, F. Kuang, Y . Yang, Z. Jin, L. Lin, and D. Povey, “Zipformer: A faster and better encoder for automatic speech recognition,” in The Twelfth International Conference on Learning Representations (ICLR), 2023

    2023

  6. [12]

    A com- parison of streaming models and data augmentation methods for robust speech recognition,

    J. Kim, M. Kumar, D. Gowda, A. Garg, and C. Kim, “A com- parison of streaming models and data augmentation methods for robust speech recognition,” in Proceedings of the IEEE Auto- matic Speech Recognition and Understanding Workshop (ASRU). IEEE, 2021, pp. 989–995

    2021

  7. [13]

    Towards fast and accurate streaming end-to-end asr,

    B. Li, S.-y. Chang, T. N. Sainath, R. Pang, Y . He, T. Strohman, and Y . Wu, “Towards fast and accurate streaming end-to-end asr,” in Proceedings of ICASSP. IEEE, 2020, pp. 6069–6073

    2020

  8. [14]

    Fastemit: Low- latency streaming asr with sequence-level emission regulariza- tion,

    J. Yu, C.-C. Chiu, B. Li, S.-y. Chang, T. N. Sainath, Y . He, A. Narayanan, W. Han, A. Gulati, Y . Wuet al., “Fastemit: Low- latency streaming asr with sequence-level emission regulariza- tion,” in 2021 IEEE international conference on acoustics, speech and signal processing (...

    2021

  9. [15]

    Delay-penalized transducer for low- latency streaming asr,

    W. Kang, Z. Yao, F. Kuang, L. Guo, X. Yang, L. Lin, P. ˙Zelasko, and D. Povey, “Delay-penalized transducer for low- latency streaming asr,” in 2023 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, 2023, pp. 1–5

    2023

  10. [18]

    Minimum latency training of se- quence transducers for streaming end-to-end speech recognition,

    Y . Shinohara and S. Watanabe, “Minimum latency training of se- quence transducers for streaming end-to-end speech recognition,” in Proceedings of Interspeech, 2022, pp. 2098–2102

    2022

  11. [19]

    Endpoint detection for streaming end- to-end multi-talker ASR,

    L. Lu, J. Li, and Y . Gong, “Endpoint detection for streaming end- to-end multi-talker ASR,” in 2022 IEEE international conference on acoustics, speech and signal processing (ICASSP) . IEEE, 2022, pp. 7312–7316

    2022

  12. [20]

    Towards accurate and real-time end-of-speech estimation,

    Y . Fan, C. Vaz, D. He, J. Heymann, V . A. Trinh, Z. Zhang, and V . Ravichandran, “Towards accurate and real-time end-of-speech estimation,” in 2023 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, 2023, pp. 1–5

    2023

  13. [21]

    Unified end-to-end speech recognition and endpointing for fast and efficient speech systems,

    S. Bijwadia, S.-y. Chang, B. Li, T. Sainath, C. Zhang, and Y . He, “Unified end-to-end speech recognition and endpointing for fast and efficient speech systems,” in 2022 IEEE Spoken Language Technology Workshop (SLT). IEEE, 2023, pp. 310–316

    2022

  14. [22]

    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,” in 2015 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, 2015, pp. 5206–5210

    2015

  15. [23]

    Switchboard: Telephone speech corpus for research and development,

    J. J. Godfrey, E. C. Holliman, and J. McDaniel, “Switchboard: Telephone speech corpus for research and development,” in 1992 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, 1992, pp. 517–520

    1992

  16. [24]

    Is the speaker done yet? Faster and more accurate end-of-utterance detection using prosody,

    L. Ferrer, E. Shriberg, and A. Stolcke, “Is the speaker done yet? Faster and more accurate end-of-utterance detection using prosody,” inProc. ICSLP, vol. 3, 2002, pp. 2061–2064

    2002

  17. [25]

    Turn-taking and backchan- nel prediction with acoustic and large language model fusion,

    J. Wang, L. Chen, A. Khare, A. Raju, P. Dheram, D. He, M. Wu, A. Stolcke, and V . Ravichandran, “Turn-taking and backchan- nel prediction with acoustic and large language model fusion,” arXiv:2401.14717, 2024, to appear in Proc. ICASSP

    2024 arXiv

  18. [26]

    Dissecting User-Perceived Latency of On-Device E2E Speech Recognition,

    Y . Shangguan, R. Prabhavalkar, H. Su, J. Mahadeokar, Y . Shi, J. Zhou, C. Wu, D. Le, O. Kalinli, C. Fuegen, and M. L. Seltzer, “Dissecting User-Perceived Latency of On-Device E2E Speech Recognition,” in Proc. INTERSPEECH 2021 – 22nd Annual Con- ference of the International Sp...

    2021

  19. [27]

    Dialogue act modeling for automatic tagging and recognition of conversational speech,

    A. Stolcke, K. Ries, N. Coccaro, E. Shriberg, R. Bates, D. Ju- rafsky, P. Taylor, R. Martin, C. V . Ess-Dykema, and M. Meteer, “Dialogue act modeling for automatic tagging and recognition of conversational speech,” Computational linguistics, vol. 26, no. 3, pp. 339–373, 2000

    2000

  20. [28]

    Metrics for polyphonic sound event detection,

    A. Mesaros, T. Heittola, and T. Virtanen, “Metrics for polyphonic sound event detection,” Applied Sciences, vol. 6, no. 6, p. 162, 2016

    2016

  21. [29]

    The third DIHARD di- arization challenge,

    N. Ryant, P. Singh, V . Krishnamohan, R. Varma, K. Church, J. D. Cieri, S. Ganapathy, and M. Liberman, “The third DIHARD di- arization challenge,” 2021

    2021

Pith tools

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