Pith. sign in

REVIEW 4 major objections 3 minor 7 cited by

NVSpeech: An Integrated and Scalable Pipeline for Human-Like Speech Modeling with Paralinguistic Vocalizations

T0 review · 4 major / 3 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read NVSpeech claims that laughter, breathing, and interjections can be transcribed and synthesized as inline text tokens in an open Mandarin pipeline.

desk verdict Plausible Mandarin paralinguistic resource, but the only text I can judge is the abstract, and the abstract omits the validation that would make the scalability claim trustworthy. read the letter →

arxiv 2508.04195 v1 pith:3CYQRHF2 submitted 2025-08-06 cs.SD cs.AIcs.LG

Huan Liao , Qinke Ni , Yuancheng Wang , Yiheng Lu , Haoyue Zhan , Pengyuan Xie , Qiang Zhang , Zhizheng Wu This is my paper
classification cs.SDcs.AIcs.LG
keywords paralinguisticvocalizationsMandarinspeechautomaticrecognitiontext-to-speechword-levelannotationzero-shotTTSexpressiveinlinetokendecoding
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

NVSpeech sets out to make paralinguistic vocalizations—laughter, breathing, sighs, and interjections like "uhm"—first-class elements of Mandarin speech technology rather than noise to be filtered out. It claims to deliver the first open, large-scale, word-level annotated pipeline for expressive Mandarin speech, built from a manually labeled set of 48,430 utterances with 18 paralinguistic categories, a paralinguistic-aware ASR that emits cues as inline tokens such as "[Laughter]", and an auto-annotated corpus of 174,179 utterances (573 hours) used to fine-tune zero-shot TTS with controllable, context-aware insertion of these cues. If the claims hold, voice assistants, dubbing, and spoken-language interfaces could transcribe and reproduce human-like nonverbal expression, and the dataset itself becomes a resource for studying affect and interaction in Mandarin.

What carries the argument

The carrying mechanism is the inline token representation of paralinguistic cues: non-verbal events are written into the text stream as bracketed tokens such as "[Laughter]", which the ASR is trained to decode and the TTS is trained to pronounce at arbitrary token positions. This single representation lets recognition and generation share a vocabulary, making word-level alignment, scalable annotation, and controllable synthesis all expressions of the same idea.

What would settle it

Take a random sample of the 174,179 auto-annotated utterances, have trained annotators mark the true positions and categories of laughter, breath, and interjections, and compare against the ASR output; per-token precision and recall near chance on any prevalent category would falsify the scalability claim.

Watch

Extended reading notes

Core claim

The paper's central claim is that paralinguistic vocalizations can be treated as ordinary, decodable tokens inside both ASR and TTS, so that a machine can read and write laughter, breath, and interjections in the same stream as words. Concretely, NVSpeech introduces 48,430 manually annotated utterances covering 18 word-level paralinguistic categories, trains a paralinguistic-aware ASR that outputs forms like "You're so funny [Laughter]", uses that ASR to auto-annotate 174,179 utterances (573 hours) with word-level alignments, and then fine-tunes zero-shot TTS models on both human- and auto-labeled data. The result, the paper argues, is the first open, large-scale, word-level annotated pipeli

Load-bearing premise

The auto-annotated 573-hour corpus is only as reliable as the paralinguistic ASR's pseudo-labels, and the abstract reports no agreement check or error analysis for those labels.

Editorial extensions

If this is right

  • ASR systems can emit non-verbal events as searchable, readable tokens instead of discarding them as noise.
  • TTS can be prompted to place laughter, breath, or an interjection at any chosen position, enabling explicit control over expressive speech.
  • The human-labeled seed data and the auto-annotation loop allow a large expressive corpus to be built without manually transcribing every paralinguistic event.
  • The 18-category word-level annotation scheme could become a reusable standard for expressive speech resources beyond Mandarin.
  • Recognition and synthesis can share one paralinguistic token inventory, simplifying the path from transcribed spontaneous speech to controllable synthetic speech.

Reading between the lines

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

  • The scalable half of the pipeline stands on the quality of the ASR's pseudo-labels; if those labels are systematically misplaced or mislabeled, the 573-hour corpus and the TTS trained on it would inherit the errors.
  • The same inline-token approach should transfer to other languages, since laughter, filled pauses, and breath have conventional textual representations across speech communities.
  • Word-level paralinguistic annotations could support downstream tasks such as emotion recognition, dialogue-act classification, and social-signal processing, not just TTS control.
  • A direct testable extension is a listening study comparing TTS trained with inline paralinguistic tokens against standard expressive TTS on naturalness, controllability, and listener engagement.
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

4 major / 3 minor

Summary. The submitted manuscript consists of an abstract for 'NVSpeech: An Integrated and Scalable Pipeline for Human-Like Speech Modeling with Paralinguistic Vocalizations' and a full text that is an unrelated paper on biomedical machine unlearning (MLLMU-Med). The abstract claims three contributions: (1) a manually annotated dataset of 48,430 Mandarin utterances with 18 word-level paralinguistic categories; (2) a paralinguistic-aware ASR that decodes vocalizations as inline tokens and is used to auto-annotate a 174,179-utterance (573-hour) corpus; and (3) zero-shot TTS fine-tuning on human- and auto-labeled data for controllable insertion of vocalizations at arbitrary token positions. No evaluation results, dataset documentation, system architecture details, or comparisons appear in the provided full text, because that text describes a different project. The central claim of offering the first open, large-scale, word-level annotated pipeline for expressive Mandarin speech modeling is therefore unverifiable from the manuscript as submitted.

Significance. If the claims in the abstract are correct, NVSpeech would be a valuable community resource: a word-level paralinguistic annotation scheme for Mandarin, a 573-hour corpus with vocalization tokens, and an integrated ASR/TTS pipeline. The idea of treating laughter, breathing, and interjections as inline decodable tokens is practically useful and the scale of the auto-annotated corpus is nontrivial. However, the manuscript provides no evidence to assess correctness: there are no agreement metrics, no ASR accuracy numbers, no TTS naturalness or controllability scores, and no comparison with existing expressive speech resources. The full-text mismatch is disqualifying for the current submission. The paper also ships no code, checkpoints, or dataset artifacts in the text, so no reproducibility claims can be checked.

major comments (4)
  1. [Full text (all sections)] The full text of the submission is a completely different paper: 'From Learning to Unlearning: Biomedical Security Protection in Multimodal Large Language Models' (MLLMU-Med). It does not contain the NVSpeech dataset construction, the paralinguistic-aware ASR, the TTS fine-tuning experiments, or any evaluation related to the abstract. As a result, every substantive claim in the abstract is unsupported by the manuscript. This is a load-bearing defect that prevents any technical review of the central contributions.
  2. [Abstract, contributions (2)-(3)] The scalability claim rests on using the paralinguistic ASR, trained on 48,430 human-annotated utterances, to auto-label 174,179 utterances that are then used for TTS fine-tuning. The abstract reports no validation of this pseudo-labeling step: no token-position accuracy, no category-level precision/recall, no agreement against held-out human labels, and no failure analysis. If the auto-labels contain systematic errors, the TTS models inherit them, undermining both the controllability and naturalness claims. This is not a minor omission; it is the load-bearing assumption for the 'scalable' half of the pipeline.
  3. [Abstract, contribution (1)] The manually annotated set of 48,430 utterances with 18 word-level paralinguistic categories is presented without any inter-annotator agreement measure, annotation protocol description, or category-level statistics. For a resource paper, the reliability of the gold labels is central; without at least Cohen's kappa or per-category counts and examples, the usefulness of the annotation set cannot be assessed.
  4. [Abstract, contribution (3)] The TTS claim of 'human-like speech synthesis' and 'context-aware insertion at arbitrary token positions' is not accompanied by any evaluation metric in the abstract or full text: no MOS, no intelligibility score, no comparison with baselines or prior expressive TTS systems, and no listening-test protocol. Even granting the pipeline design, the claim is empirically unsupported as presented.
minor comments (3)
  1. [Abstract] There is a typo in 'paralingustic cues' (should be 'paralinguistic'). The term 'first large-scale Chinese dataset' should be clarified as 'Mandarin' to match the title and avoid ambiguity with other Chinese dialects.
  2. [Abstract] The dataset webpage is mentioned but no license, access conditions, or persistent identifier are given. For a claimed open dataset, this should be specified.
  3. [Abstract] The phrase 'zero-shot TTS models' is vague; the underlying TTS architecture (e.g., VALL-E-style or diffusion-based) and the training corpus should be identified for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the abstract describes a bootstrap pseudo-labeling pipeline; no derivation reduces to its inputs, and the supplied full text is an unrelated manuscript.

full rationale

The submitted abstract (arXiv:2508.04195) makes resource and pipeline claims: 48,430 manually labeled utterances, a paralinguistic-aware ASR trained on them to auto-annotate 174,179 utterances, and TTS fine-tuning on both human- and auto-labeled data. None of these steps is a formal derivation from a premise to a conclusion, so the definitional or fitted-input patterns of circularity do not apply. The auto-annotation step uses the ASR trained on manual labels to produce pseudo-labels, but this is a standard bootstrap/pseudo-labeling design; it is not a 'prediction' that is forced by construction, and no equation or definition identifies the auto-labeled output with the human-labeled input. The weaker concern—that the abstract reports no held-out validation of the auto-label quality—is a correctness and validation risk, not a circularity. The supplied full text is actually a different manuscript on biomedical machine unlearning (arXiv:2508.04192), so the actual evaluation sections of NVSpeech cannot be inspected; this mismatch prevents any further circularity analysis but does not itself constitute a circular step. There are no load-bearing self-citations, no imported uniqueness theorems, and no renamed empirical pattern in the provided material. The honest finding is therefore no significant circularity.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The paper's load-bearing assumptions are practical rather than mathematical: the token representation of vocalizations, the quality of the auto-labeling loop, and the adequacy of the human seed set. There are no fitted numerical parameters in the abstract besides the hand-chosen 18-category taxonomy. Because the full text supplied is a different manuscript, the ledger is reconstructed from the abstract, and modeling choices (loss weights, thresholds, model sizes) are unavailable.

free parameters (1)
  • 18 paralinguistic categories
    Hand-chosen taxonomy of vocalization types that defines the annotation task, the ASR output space, and the TTS control vocabulary. Every downstream claim depends on this choice, and the abstract provides no criterion (e.g., inter-annotator agreement) for its adequacy.
assumptions (3)
  • domain assumption Paralinguistic events are faithfully representable as discrete inline tokens (e.g., [Laughter]) placed word-level in the transcript.
    The entire ASR decoding and TTS insertion scheme rests on this representation, introduced in abstract contribution (2) with the 'You're so funny [Laughter]' example and used in (3) for token-position insertion.
  • domain assumption Auto-labels from the paralinguistic ASR are of sufficient quality to serve as training targets for TTS.
    The scaling claim, moving from contribution (2) to (3), uses the 174,179-utterance auto-labeled corpus to fine-tune TTS; the abstract gives no quality check for these labels.
  • domain assumption Human annotation of 48,430 utterances generalizes across speakers, channels, and the 18 categories so the trained recognizer transfers to the larger corpus.
    Contribution (2) assumes coverage and consistency of the seed set suffice to train an ASR that works on 573 hours of new audio; no coverage or agreement statistics are reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of NVSpeech: An Integrated and Scalable Pipeline for Human-Like Speech Modeling with Paralinguistic Vocalizations." pith.science (2026). https://pith.science/paper/3CYQRHF2

@misc{pith2026250804195,
  author       = {Pith},
  title        = {Pith review of: NVSpeech: An Integrated and Scalable Pipeline for Human-Like Speech Modeling with Paralinguistic Vocalizations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3CYQRHF2}},
  note         = {Machine review of arXiv:2508.04195}
}
read the original abstract

Paralinguistic vocalizations-including non-verbal sounds like laughter and breathing, as well as lexicalized interjections such as "uhm" and "oh"-are integral to natural spoken communication. Despite their importance in conveying affect, intent, and interactional cues, such cues remain largely overlooked in conventional automatic speech recognition (ASR) and text-to-speech (TTS) systems. We present NVSpeech, an integrated and scalable pipeline that bridges the recognition and synthesis of paralinguistic vocalizations, encompassing dataset construction, ASR modeling, and controllable TTS. (1) We introduce a manually annotated dataset of 48,430 human-spoken utterances with 18 word-level paralinguistic categories. (2) We develop the paralinguistic-aware ASR model, which treats paralinguistic cues as inline decodable tokens (e.g., "You're so funny [Laughter]"), enabling joint lexical and non-verbal transcription. This model is then used to automatically annotate a large corpus, the first large-scale Chinese dataset of 174,179 utterances (573 hours) with word-level alignment and paralingustic cues. (3) We finetune zero-shot TTS models on both human- and auto-labeled data to enable explicit control over paralinguistic vocalizations, allowing context-aware insertion at arbitrary token positions for human-like speech synthesis. By unifying the recognition and generation of paralinguistic vocalizations, NVSpeech offers the first open, large-scale, word-level annotated pipeline for expressive speech modeling in Mandarin, integrating recognition and synthesis in a scalable and controllable manner. Dataset and audio demos are available at https://nvspeech170k.github.io/.

Discussion (0). Sign in to comment.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. NVBench: A Benchmark for Speech Synthesis with Non-Verbal Vocalizations

    cs.SD 2026-04 unverdicted novelty 7.0 of 10

    NVBench provides a standardized bilingual benchmark and evaluation protocol for assessing non-verbal vocalization generation, placement, and salience in text-to-speech systems.

  2. Transcription Policy as a Latent Variable: Activating Controllable Verbatim ASR with Word-Level Timing

    cs.CL 2026-07 conditional novelty 6.0 of 10

    Mode-tag conditioning on paired verbatim/intended data makes Whisper produce either verbatim or intended transcripts on demand, with cross-lingual disfluency control and improved word timestamps.

  3. Towards Fine-Grained Multi-Dimensional Speech Understanding: Data Pipeline, Benchmark, and Model

    eess.AS 2026-05 unverdicted novelty 6.0 of 10

    A data pipeline, 14-dimension benchmark, and decoupled fine-tuning model are presented to advance fine-grained multi-dimensional speech understanding in LLMs.

  4. TTS-PRISM: A Perceptual Reasoning and Interpretable Speech Model for Fine-Grained Diagnosis

    cs.CL 2026-04 unverdicted novelty 6.0 of 10

    TTS-PRISM defines a 12-dimensional perceptual schema, builds a targeted diagnostic dataset via adversarial synthesis and expert labels, and tunes an end-to-end model that outperforms generalist LLMs in human alignment...

  5. OmniVoice: Towards Omnilingual Zero-Shot Text-to-Speech with Diffusion Language Models

    cs.CL 2026-04 unverdicted novelty 6.0 of 10

    OmniVoice introduces a diffusion language model-style non-autoregressive TTS system that directly maps text to multi-codebook acoustic tokens, scaling zero-shot synthesis to over 600 languages with SOTA results on mul...

  6. TriA Pipeline: A Large-Scale Automatic Audio Annotation Pipeline For Audio Classification In Specific Scenarios

    eess.AS 2026-07 conditional novelty 4.0 of 10

    An automatic audio annotation pipeline using BEATs and CLAP filtering produces a 2130-hour dataset that yields 3.97% average accuracy gains on three domestic audio classification tasks.

  7. Beyond Words: Towards Effective Modeling of Non-Verbal Vocalizations in ASR

    eess.AS 2026-07 unverdicted novelty 4.0 of 10

    Three data-centric strategies are studied to improve rare non-verbal vocalization recognition in ASR while preserving lexical accuracy.

Reference graph

Works this paper leans on

38 extracted references · 21 canonical work pages · cited by 7 Pith papers

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    An, K.; Chen, Q.; Deng, C.; Du, Z.; Gao, C.; Gao, Z.; Gu, Y.; He, T.; Hu, H.; Hu, K.; et al. 2024. Funaudiollm: Voice understanding and generation foundation models for natural interaction between humans and llms. arXiv preprint arXiv:2407.04051

    arXiv 2024

  4. [4]

    Chaudhury, R.; Godbole, M.; Garg, A.; and Seo, J. H. 2024. Humane Speech Synthesis through Zero-Shot Emotion and Disfluency Generation. arXiv preprint arXiv:2404.01339

    work page Pith review arXiv 2024

  5. [5]

    Chen, S.; Wu, Y.; Wang, C.; Liu, S.; Tompkins, D.; Chen, Z.; and Wei, F. 2022. Beats: Audio pre-training with acoustic tokenizers. arXiv preprint arXiv:2212.09058

    arXiv 2022

  6. [6]

    Chu, Y.; Xu, J.; Zhou, X.; Yang, Q.; Zhang, S.; Yan, Z.; Zhou, C.; and Zhou, J. 2023. Qwen-audio: Advancing universal audio understanding via unified large-scale audio-language models. arXiv preprint arXiv:2311.07919

    arXiv 2023

  7. [7]

    DeepSeek-AI; Guo, D.; Yang, D.; Zhang, H.; Song, J.; et al. 2025. DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning. arXiv:2501.12948

    arXiv 2025

  8. [8]

    Deng, Y.-C.; Liao, Y.-F.; Wang, Y.-R.; and Chen, S.-H. 2023. Toward enriched decoding of mandarin spontaneous speech. Speech Communication, 154: 102983

    work page 2023

Show all 38 references
  1. [9]

    Du, Z.; Wang, Y.; Chen, Q.; Shi, X.; Lv, X.; Zhao, T.; Gao, Z.; Yang, Y.; Gao, C.; Wang, H.; et al. 2024. Cosyvoice 2: Scalable streaming speech synthesis with large language models. arXiv preprint arXiv:2412.10117

    2024 arXiv

  2. [10]

    Gao, Z.; Li, Z.; Wang, J.; Luo, H.; Shi, X.; Chen, M.; Li, Y.; Zuo, L.; Du, Z.; Xiao, Z.; et al. 2023. Funasr: A fundamental end-to-end speech recognition toolkit. arXiv 2023. arXiv preprint arXiv:2305.11013

    2023 arXiv

  3. [11]

    Gao, Z.; Zhang, S.; McLoughlin, I.; and Yan, Z. 2022. Paraformer: Fast and accurate parallel transformer for non-autoregressive end-to-end speech recognition. arXiv preprint arXiv:2206.08317

    2022 arXiv

  4. [12]

    F.; Ellis, D

    Gemmeke, J. F.; Ellis, D. P. W.; Freedman, D.; Jansen, A.; Lawrence, W.; Moore, R. C.; Plakal, M.; and Ritter, M. 2017. Audio Set: An ontology and human-labeled dataset for audio events. In Proc. IEEE ICASSP 2017. New Orleans, LA

    2017

  5. [13]

    Gong, Y.; Yu, J.; and Glass, J. 2022. Vocalsound: A dataset for improving human vocal sounds recognition. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 151--155. IEEE

    2022

  6. [14]

    Graves, A.; Fern \'a ndez, S.; Gomez, F.; and Schmidhuber, J. 2006. Connectionist temporal classification: labelling unsegmented sequence data with recurrent neural networks. In Proceedings of the 23rd international conference on Machine learning, 369--376

    2006

  7. [15]

    Guo, H.-H.; Hu, Y.; Liu, K.; Shen, F.-Y.; Tang, X.; Wu, Y.-C.; Xie, F.-L.; Xie, K.; and Xu, K.-T. 2024. Fireredtts: A foundation text-to-speech framework for industry-level generative speech applications. arXiv preprint arXiv:2409.03283

    2024 arXiv

  8. [16]

    Gupta, R.; Audhkhasi, K.; Lee, S.; and Narayanan, S. 2016. Detecting paralinguistic events in audio stream using context in features and probabilistic decisions. Computer speech & language, 36: 72--92

    2016

  9. [17]

    He, H.; Shang, Z.; Wang, C.; Li, X.; Gu, Y.; Hua, H.; Liu, L.; Yang, C.; Li, J.; Shi, P.; et al. 2025. Emilia: A Large-Scale, Extensive, Multilingual, and Diverse Dataset for Speech Generation. arXiv preprint arXiv:2501.15907

    2025

  10. [18]

    E.; Thakker, M.; Yang, H.; Zhu, Z.; Tang, M.; Li, C.; Tsai, C.-H.; Xiao, Z.; et al

    Kanda, N.; Wang, X.; Eskimez, S. E.; Thakker, M.; Yang, H.; Zhu, Z.; Tang, M.; Li, C.; Tsai, C.-H.; Xiao, Z.; et al. 2024. Making flow-matching-based zero-shot text-to-speech laugh as you like. arXiv preprint arXiv:2402.07383

    2024 arXiv

  11. [19]

    S.; and Aslin, R

    Kidd, C.; White, K. S.; and Aslin, R. N. 2011. Toddlers use speech disfluencies to predict speakers’ referential intentions. Developmental science, 14(4): 925--934

    2011

  12. [20]

    Kong, Q.; Cao, Y.; Iqbal, T.; Wang, Y.; Wang, W.; and Plumbley, M. D. 2020. Panns: Large-scale pretrained audio neural networks for audio pattern recognition. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 28: 2880--2894

    2020

  13. [21]

    M.; and Nass, C

    Lee, K. M.; and Nass, C. 2003. Designing social presence of social actors in human computer interaction. In Proceedings of the SIGCHI conference on Human factors in computing systems, 289--296

    2003

  14. [22]

    E.; Rohde, H.; and Corley, M

    Loy, J. E.; Rohde, H.; and Corley, M. 2017. Effects of disfluency in online interpretation of deception. Cognitive Science, 41: 1434--1456

    2017

  15. [23]

    A.; Hsu, W.-N.; d'Avirro, A.; Shi, B.; Gat, I.; Fazel-Zarani, M.; Remez, T.; Copet, J.; Synnaeve, G.; Hassid, M.; et al

    Nguyen, T. A.; Hsu, W.-N.; d'Avirro, A.; Shi, B.; Gat, I.; Fazel-Zarani, M.; Remez, T.; Copet, J.; Synnaeve, G.; Hassid, M.; et al. 2023. Expresso: A benchmark and analysis of discrete expressive speech resynthesis. arXiv preprint arXiv:2308.05725

    2023 arXiv

  16. [24]

    Polychroniou, A.; Salamin, H.; and Vinciarelli, A. 2014. The SSPNet-Mobile Corpus: Social Signal Processing Over Mobile Phones. In LREC, 1492--1498

    2014

  17. [25]

    W.; Xu, T.; Brockman, G.; McLeavey, C.; and Sutskever, I

    Radford, A.; Kim, J. W.; Xu, T.; Brockman, G.; McLeavey, C.; and Sutskever, I. 2023. Robust speech recognition via large-scale weak supervision. In International conference on machine learning, 28492--28518. PMLR

    2023

  18. [26]

    M.; Li, G.; and Du, C

    Rashid, M. M.; Li, G.; and Du, C. 2023. Nonspeech7k dataset: Classification and analysis of human non-speech sound. IET Signal Processing, 17(6): e12233

    2023

  19. [27]

    Rennie, G.; Perepelkina, O.; and Vinciarelli, A. 2022. Which Model is Best: Comparing Methods and Metrics for Automatic Laughter Detection in a Naturalistic Conversational Dataset. In INTERSPEECH, 4008--4012

    2022

  20. [28]

    Saeki, T.; Xin, D.; Nakata, W.; Koriyama, T.; Takamichi, S.; and Saruwatari, H. 2022. Utmos: Utokyo-sarulab system for voicemos challenge 2022. arXiv preprint arXiv:2204.02152

    2022 arXiv

  21. [29]

    Salamin, H.; Polychroniou, A.; and Vinciarelli, A. 2013. Automatic detection of laughter and fillers in spontaneous mobile phone conversations. In 2013 IEEE International Conference on Systems, Man, and Cybernetics, 4282--4287. IEEE

    2013

  22. [30]

    San Segundo, E.; and Mompean, J. A. 2017. A simplified vocal profile analysis protocol for the assessment of voice quality and speaker similarity. Journal of Voice, 31(5): 644--e11

    2017

  23. [31]

    Tan, X.; Qin, T.; Soong, F.; and Liu, T.-Y. 2021. A survey on neural speech synthesis. arXiv preprint arXiv:2106.15561

    2021 arXiv

  24. [32]

    Tseng, S.-C. 2003. Taxonomy of spontaneous speech phenomena in Mandarin conversation. In Proc. of ISCA & IEEE Workshop on Spontaneous Speech Processing and Recognition, 23--26

    2003

  25. [33]

    Tseng, S.-C. 2013. Lexical coverage in Taiwan Mandarin conversation. In International Journal of Computational Linguistics & Chinese Language Processing, Volume 18, Number 1, March 2013

    2013

  26. [34]

    Wang, K.; and Herremans, D. 2024. DisfluencySpeech--Single-Speaker Conversational Speech Dataset with Paralanguage. arXiv preprint arXiv:2406.08820

    2024 arXiv

  27. [35]

    Ward, N. 2006. Non-lexical conversational sounds in American English. Pragmatics & Cognition, 14(1): 129--182

    2006

  28. [36]

    E.; Thakker, M.; Tompkins, D.; Tsai, C.-H.; Li, C.; Xiao, Z.; Zhao, S.; Li, J.; et al

    Wu, H.; Wang, X.; Eskimez, S. E.; Thakker, M.; Tompkins, D.; Tsai, C.-H.; Li, C.; Xiao, Z.; Zhao, S.; Li, J.; et al. 2024. Laugh Now Cry Later: Controlling Time-Varying Emotional States of Flow-Matching-Based Zero-Shot Text-To-Speech. In 2024 IEEE Spoken Language Technology Wo...

    2024

  29. [37]

    Yang, Z.; Chen, Y.; Luo, L.; Yang, R.; Ye, L.; Cheng, G.; Xu, J.; Jin, Y.; Zhang, Q.; Zhang, P.; et al. 2022. Open source magicdata-ramc: A rich annotated mandarin conversational (ramc) speech dataset. arXiv preprint arXiv:2203.16844

    2022 arXiv

  30. [38]

    Zhang, H.; Yu, X.; and Lin, Y. 2023. NSV-TTS: Non-speech vocalization modeling and transfer in emotional text-to-speech. In ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 1--5. IEEE

    2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.