The DOI matches its Crossref record; 28 of the 36 references it cites were found in a registry or an index; 18 of the 29 sources its claims cite were checked for support.
Vocalsound: A Dataset for Improving Human Vocal Sounds Recognition
Open accessAbstract
A VocalSound dataset consisting of over 21,000 crowdsourced recordings of laughter, sighs, coughs, throat clearing, sneezes, and sniffs from 3,365 unique subjects is created to support research on building robust and accurate vocal sound recognition.
Overview
CoChat Summary
TL;DR
VocalSound is a crowdsourced dataset of over 21,000 human non-speech vocal sound recordings from 3,365 speakers, designed to improve vocal sound recognition. Experiments show 90.5 ± 0.2% accuracy on a 6-class task and a 41.9% relative average precision improvement when adding VocalSound to FSD50K for real-sound evaluation.
Key findings
- The dataset contains over 21,000 recordings of six classes: laughter, sighs, coughs, throat clearing, sneezes, and sniffs, collected from 3,365 unique subjects across 60 countries.
- In a 6-class classification experiment, the model achieved 90.5 ± 0.2% accuracy on the evaluation set and 90.1 ± 0.2% on the validation set.
- Performance varied by speaker group: age 18–25 was 91.5 ± 0.3%, age 26–48 was 90.1 ± 0.2%, age 49–80 was 90.9 ± 1.6%; male was 89.2 ± 0.5% and female was 91.9 ± 0.1%.
- Adding VocalSound to FSD50K improved vocal sound recognition on FSD50K evaluation by 31.8% relative F1 and 41.9% relative AP; the improvement was statistically significant with McNemar’s test, p < 0.05.
Methods
- Crowdsourced data collection via Amazon Mechanical Turk: each HIT asked for demographics/health info and recordings of six vocal sounds; recordings were anonymized and filtered by duration, Google Speech API transcription checks, and a pretrained model check.
- The dataset was split speaker-independently into training, validation, and evaluation sets with 15,570 (74%), 1,860 (9%), and 3,594 (17%) samples, respectively.
- Baseline models used an EfficientNet-B0 audio classifier with log-Mel fbank features, Adam (1e-4), batch size 100, cross-entropy loss, 50 epochs, SpecAugment, and 3 repeated runs; the second experiment combined VocalSound with FSD50K and used weighted random sampling.
Limitations
- The paper states one limitation explicitly: the audio samples are “not produced spontaneously but acted by the subjects.”
- The text also makes evident that the dataset may not fully represent spontaneous real-world vocalizations because recordings were crowdsourced and performance is evaluated partly by transfer to FSD50K real sounds; this is an inference from the study design, not stated verbatim in the paper.
- Not stated in the paper.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
9 Valid · 5 Not valid · 2 Not checked · 36 Own results and interpretations
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“Since FSD50K only contains 4 vocal sound classes”
What the sources say
- Disputedcites [8] [8] E. Fonseca, X. Favory, J. Pons, F. Font, and X. Serra, 'Fsd50k: an open dataset of human-labeled sound events,' arXiv preprint arXiv:2010.00475 , 2020. - sourceExperimental — accuracy not yet measured
The Baseline Systems section (p. 15) identifies more than four FSD50K classes involving human vocal sounds: it distinguishes spoken, sung, screamed, yelled, and whispered speech, and also names laughter, gasping, and crying classes. The paper therefore contradicts the claim that FSD50K contains only four vocal sound classes. (LLM, conf=0.85)
Checked by a reasoning model — may be wrong.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Disputed
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“As a consequence, state-of-the-art audio event classification models may not perform well in detecting human vocal sounds.”
- Also stated in: § 1. INTRODUCTION · plain
“Moreover, it has been found that a generic audio event classification model trained with existing datasets such as AudioSet [4] does not perform well in classifying human vocal sounds, e.g., the average precision of state-of-the-art models on cough and sneeze classes are only around 0.5 on the AudioSet evaluation set [5, 6].”
What the sources say
- Not checked — no full text of the sourcecites [5] [5] Q. Kong, Y. Cao, T. Iqbal, Y. Wang, W. Wang, and M. D. Plumbley, 'Panns: Large-scale pretrained audio neural networks for audio pattern recognition,' IEEE/ACM Transactions on Audio, Speech, and Language Processing , 2020. - source
- Partly supportedcites [6] [6] Y. Gong, Y.-A. Chung, and J. Glass, 'Psla: Improving audio tagging with pretraining, sampling, labeling, and aggregation,' IEEE/ACM Transactions on Audio, Speech, and Language Processing , 2021. - sourceExperimental — accuracy not yet measured
In V. Label Enhancement (p. 7), the paper reports low performance for male, female, and child speech classes and says class-specific performance patterns are similar across model architectures. But the Fig. 7 caption (p. 7) says the broader “Speech” class has much higher average precision, so the evidence does not extend to human vocal sounds generally. (LLM, conf=0.87)
plainstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 1. INTRODUCTION · plain
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“the average precision of state-of-the-art models on cough and sneeze classes are only around 0.5 on the AudioSet evaluation set [5, 6].”
What the sources say
- Not checked — no full text of the sourcecites [5] [5] Q. Kong, Y. Cao, T. Iqbal, Y. Wang, W. Wang, and M. D. Plumbley, 'Panns: Large-scale pretrained audio neural networks for audio pattern recognition,' IEEE/ACM Transactions on Audio, Speech, and Language Processing , 2020. - source
- Could not verifycites [6] [6] Y. Gong, Y.-A. Chung, and J. Glass, 'Psla: Improving audio tagging with pretraining, sampling, labeling, and aggregation,' IEEE/ACM Transactions on Audio, Speech, and Language Processing , 2021. - sourceExperimental — accuracy not yet measured
The Introduction (p. 1) reports an overall AudioSet mAP of 0.474, and the Figure 7 caption (p. 7) discusses class-wise AP, but neither gives AP values for cough or sneeze. Searches for those classes and a class-wise value near 0.5 found no passage that verifies the claim. (LLM, conf=0.72)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Due to the difficulty of annotating YouTube videos with a large sound vocabulary, the noisy label issue has been found in AudioSet [6, 9, 10]”
What the sources say
- Supportedcites [6] [6] Y. Gong, Y.-A. Chung, and J. Glass, 'Psla: Improving audio tagging with pretraining, sampling, labeling, and aggregation,' IEEE/ACM Transactions on Audio, Speech, and Language Processing , 2021. - source
The Label Enhancement section (pp. 7–8) explicitly examines noisy labels in AudioSet and reports prevalent false-negative annotations, including in speech classes. This supports the cited claim that AudioSet has a noisy-label issue; the paper does not need to establish the sentence’s introductory explanation as a cause. (LLM, conf=0.94)
- Could not verifycites [9] [9] E. Fonseca, S. Hershey, M. Plakal, D. P. Ellis, A. Jansen, and R. C. Moore, 'Addressing missing labels in largescale sound event recognition using a teacher-student framework with loss masking,' IEEE Signal Processing Letters , 2020. - sourceExperimental — accuracy not yet measured
Our copy mentions AudioSet only twice and never in connection with label noise: Sec. 2 (p. 1) cites it (Gemmeke et al. 2017) as an open-source dataset that 'further boosted research in this domain', and Sec. 3.2 (p. 4) notes Flusense is 'a subset of Google's Audioset dataset'. The paper's own label-noise discussion concerns a different source entirely — RT-PCR test sensitivity (~70% at 95% specificity) making its COVID ground truth labels possibly wrong (Sec. 4.3, p. 5; Sec. 2, p. 2), plus silence at segment start/end producing noisy labels in random crops (Sec. 6.4, p. 7). Searched: 'AudioSet noisy labels annotation YouTube videos', 'label noise weak labels dataset', 'Flusense AudioSet subset labels weakness', 'limitations of dataset labeling errors false positives' — no passage attributes a noisy-label issue to AudioSet, and none contradicts it either. (LLM, conf=0.70)
- Supportedcites [10] [10] A. Shah, A. Kumar, A. G. Hauptmann, and B. Raj, 'A closer look at weak label learning for audio events,' arXiv preprint arXiv:1804.09288 , 2018. - source
The Introduction (p. 2) states that label corruption noise arises from actual wrong labeling, that this worsens at large scale with many audio events, and that for AudioSet — though manually labeled — label quality 'has been roughly estimated to be less than 90% for several events.' Section IV.B 'Corrupted Labels Noise' (p. 5) repeats this, noting the AudioSet authors themselves sampled 10 recordings per event and reported confidence below 90% for several events, and that YouTube-quality audio and hard-to-interpret sound events make labeling difficult. This covers both parts of the citing sentence: the annotation difficulty and the presence of noisy labels in AudioSet. (LLM, conf=0.90)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“Also, limited by the data acquisition scheme, they do not provide speaker information such as age, gender, native language, etc.”
What the sources say
- Not checked — no full text of the sourcecites [4] [4] J. F. Gemmeke, D. P. Ellis, D. Freedman, A. Jansen, W. Lawrence, R. C. Moore, M. Plakal, and M. Ritter, 'Audio Set: An ontology and human-labeled dataset for audio events,' in ICASSP , 2017. - source
- Could not verifycites [8] [8] E. Fonseca, X. Favory, J. Pons, F. Font, and X. Serra, 'Fsd50k: an open dataset of human-labeled sound events,' arXiv preprint arXiv:2010.00475 , 2020. - sourceExperimental — accuracy not yet measured
The Data acquisition section (p. 4) describes Freesound titles, tags, and descriptions, and the FSD50K and AudioSet comparison (p. 13) describes their sound-event annotations. Neither says whether the datasets provide speaker age, gender, or native-language information, or attributes any absence to their acquisition schemes. Searches for speaker demographics and those specific attributes found no decisive passage. (LLM, conf=0.72)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
24 more claims · 1 with their cited sources checked
Create a free account to see all 52 claims and to verify the 2 references still open.
- It may state two separate claims as one. 13
- Probably not a substantial claim. 15
Citations
Cites
Adam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946 · cited 1×
“For both experiments, we train the model using an Adam optimizer [22], an initial learning rate of 1e-4, a batch size of 100, and cross-entropy loss for 50 epochs and select the best model using the development set and evaluate the model on the evaluation set.”
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
Tan, M., & Le, Q. V. (2019) · arXiv · doi:10.48550/arxiv.1905.11946 · Cited by 17,072
SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition
Park, D. S., Chan, W., Zhang, Y., Chiu, C.-C., Zoph, B., Cubuk, E. D., & Le, Q. V. (2019) · Interspeech 2019 · doi:10.21437/interspeech.2019-2680 · Cited by 3,558 · cited 2×
“In addition to balanced sampling, we also apply SpecAugment [23] and random time shift to alleviate the class-imbalance issue.”
Audio Set: An ontology and human-labeled dataset for audio events
Gemmeke, J. F., Ellis, D. P. W., Freedman, D., Jansen, A., Lawrence, W., Moore, R. C., Plakal, M., & Ritter, M. (2017) · 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp.2017.7952261 · Cited by 3,134 · cited 4×
“Potential reasons include the fact that corpora such as ESC-50 [7], FSD50K [8], and AudioSet [4] have a relatively small number of human vocal sound samples (summarized in Table 1) and the AudioSet annotation quality for these sounds may be lacking due to the challenge of annotating with a large sound vocabulary [6, 9…”
InfluentialESC: Dataset for Environmental Sound Classification
Piczak, K. J. (2015) · Proceedings of the 23rd ACM international conference on Multimedia · doi:10.1145/2733373.2806390 · Cited by 1,397 · cited 4×
“Potential reasons include the fact that corpora such as ESC-50 [7], FSD50K [8], and AudioSet [4] have a relatively small number of human vocal sound samples (summarized in Table 1) and the AudioSet annotation quality for these sounds may be lacking due to the challenge of annotating with a large sound vocabulary [6, 9…”
InfluentialPANNs: Large-Scale Pretrained Audio Neural Networks for Audio Pattern Recognition
Kong, Q., Cao, Y., Iqbal, T., Wang, Y., Wang, W., & Plumbley, M. D. (2020) · IEEE/ACM Transactions on Audio, Speech, and Language Processing · doi:10.1109/taslp.2020.3030497 · Cited by 1,235 · cited 1×
“Even the state-of-the-art audio classification models in [6, 5] cannot achieve satisfactory results for the vocal sound classes, e.”
AI4COVID-19: AI enabled preliminary diagnosis for COVID-19 from cough samples via an app
Imran, A., Posokhova, I., Qureshi, H. N., Masood, U., Riaz, M. S., Ali, K., John, C. N., Hussain, M. I., & Nabeel, M. (2020) · Informatics in Medicine Unlocked · doi:10.1016/j.imu.2020.100378 · Cited by 621 · cited 1×
“Recently, there are a few efforts to collect cough samples for building COVID-19 classification models [12, 13, 14, 15, 16, 17, 18].”
COVID-19 Artificial Intelligence Diagnosis Using Only Cough Recordings
Laguarta, J., Hueto, F., & Subirana, B. (2020) · IEEE Open Journal of Engineering in Medicine and Biology · doi:10.1109/ojemb.2020.3026928 · Cited by 589 · cited 1×
“Recently, there are a few efforts to collect cough samples for building COVID-19 classification models [12, 13, 14, 15, 16, 17, 18].”
FSD50K: An Open Dataset of Human-Labeled Sound Events
Fonseca, E., Favory, X., Pons, J., Font, F., & Serra, X. (2020) · arXiv · doi:10.48550/arxiv.2010.00475 · Cited by 462
FSD50K: An Open Dataset of Human-Labeled Sound Events
Fonseca, E., Favory, X., Pons, J., Font, F., & Serra, X. (2022) · IEEE/ACM Transactions on Audio, Speech, and Language Processing · doi:10.1109/taslp.2021.3133208 · Cited by 449 · cited 4×
“Potential reasons include the fact that corpora such as ESC-50 [7], FSD50K [8], and AudioSet [4] have a relatively small number of human vocal sound samples (summarized in Table 1) and the AudioSet annotation quality for these sounds may be lacking due to the challenge of annotating with a large sound vocabulary [6, 9…”
Influential
Research works printed but not linked: 1
- kim, 2018, Vocal imitation set: a dataset of vocally imitated sound events using the audioset ontology, DCAS
Cited by
CLAP Learning Audio Concepts from Natural Language Supervision
Elizalde, B., Deshmukh, S., Ismail, M. A., & Wang, H. (2023) · ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp49357.2023.10095889 · Cited by 449
Qwen-Audio: Advancing Universal Audio Understanding via Unified Large-Scale Audio-Language Models
Chu, Y., Xu, J., Zhou, X., Yang, Q., Zhang, S., Yan, Z., Zhou, C., & Zhou, J. (2023) · arXiv · doi:10.48550/arxiv.2311.07919 · Cited by 61
Nonspeech7k dataset: Classification and analysis of human non‐speech sound
Rashid, M. M., Li, G., & Du, C. (2023) · IET Signal Processing · doi:10.1049/sil2.12233 · Cited by 26
A Survey on Speech Large Language Models for Understanding
Peng, J., Wang, Y., Li, B., Guo, Y., Wang, H., Fang, Y., Xi, Y., Li, H., Li, X., Zhang, K., et al. (2025) · IEEE Journal of Selected Topics in Signal Processing · doi:10.1109/jstsp.2025.3640535 · Cited by 149
On Out-of-Distribution Detection for Audio with Deep Nearest Neighbors
Bukhsh, Z. A., & Saeed, A. (2023) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49357.2023.10094846 · Cited by 15
“We use MSCW (Micro-EN) [10] and Vocalsound [11] as indistribution datasets, and they also act as OOD for each other.”
MethodInfluential
Metadata
- Authors
- Gong, Y., Yu, J., & Glass, J.
- Affiliations
- MIT Computer Science and Artificial Intelligence Laboratory (US)
- Signify Research,Cambridge,MA,USA,02142
- Venue
- ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
- Publisher
- IEEE
- Published
- 2022-05-23
- Pages
- 151-155
- ISSN
- 1520-6149 (ISSN-L)
- OpenAlex
- W4225299287
- Cited by
- 43
- References
- 36
- Citations by year
- 2026: 11 · 2025: 15 · 2024: 8 · 2023: 7 · 2022: 2
- FWCI
- 5.69
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- IEEE International Conference on Acoustics Speech and Signal Processing (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University) (cc-by)
- Topics
- Physical Sciences › Computer Science › Signal Processing › Music and Audio Processing
- Physical Sciences › Computer Science › Artificial Intelligence › Speech Recognition and Synthesis
- Physical Sciences › Computer Science › Signal Processing › Speech and Audio Processing
- Keywords
- sound event classification, laughter, cough, sneezing, throat clearing, sighs, speaker metadata, health monitoring
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-09, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-09 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Vocalsound: A Dataset for Improving Human Vocal Sounds Recognition. https://app.cochat.ai/papers/10.1109/icassp43922.2022.9746828
- The paper
- Gong, Y., Yu, J., & Glass, J. (2022). Vocalsound: A Dataset for Improving Human Vocal Sounds Recognition. ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). https://doi.org/10.1109/icassp43922.2022.9746828
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.