Re-annotation of cough events in the AMI corpus

Damon Berry; David Dorran; Paul Leamy; Ted Burke

arxiv: 1906.11509 · v1 · pith:DE2XJUKLnew · submitted 2019-06-27 · 📡 eess.AS · eess.SP

Re-annotation of cough events in the AMI corpus

Paul Leamy , Ted Burke , Damon Berry , David Dorran This is my paper

Pith reviewed 2026-05-25 14:14 UTC · model grok-4.3

classification 📡 eess.AS eess.SP

keywords cough detectionaudio event detectionAMI corpusre-annotationsound databasemachine learningpublic dataset

0 comments

The pith

Re-annotating cough locations in the AMI corpus yields 1369 events released with an open annotation tool.

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

The paper establishes a publicly available re-annotated set of cough events drawn from the Augmented Multi-party Interaction corpus. The authors built a MATLAB GUI that imports audio files and lets users mark cough onsets and offsets quickly, following a consistent procedure across the recordings. This produces a dataset of 1369 individual cough events that can be downloaded along with the tool itself. The work targets the shortage of reliable, shareable cough recordings that machine-learning systems need for training and testing automatic cough detectors, given the ethical barriers to creating fresh recordings.

Core claim

A reusable re-annotation workflow implemented through a custom MATLAB GUI produces a corrected set of cough location labels for the AMI corpus containing 1369 individual cough events, with both the labels and the annotation tool released for public download and use in audio event detection research.

What carries the argument

A purpose-built MATLAB GUI for importing audio signals and marking cough event locations according to a defined reusable methodology.

If this is right

Developers can train and benchmark new machine-learning models for cough detection directly on the released 1369-event set.
The public labels remove the need for each research group to repeat the annotation effort on the same source material.
The released GUI and workflow can be applied to other audio corpora that contain similar non-speech events.
Standardized evaluation of detection algorithms becomes possible because all groups start from the same annotated events.

Where Pith is reading between the lines

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

The dataset may support studies that combine cough counts with meeting metadata already present in the AMI recordings.
Releasing the tool lowers the barrier for groups that want to correct annotations in other meeting or conversation corpora.
If the labels prove stable under further review, they could serve as a baseline for testing automated pre-labeling methods before human correction.

Load-bearing premise

The single manual pass through the recordings by the authors correctly identifies genuine cough events without independent checks from other listeners or clinicians.

What would settle it

An independent listening test on a random sample of the annotated segments that reports a substantial fraction of the 1369 events as either non-coughs or missed coughs.

Figures

Figures reproduced from arXiv: 1906.11509 by Damon Berry, David Dorran, Paul Leamy, Ted Burke.

**Figure 1.** Figure 1: Recordings were captured from omni-directional microphone arrays (near and far mounted) and headset and lapel microphones - Dashed insert: Closeup of omni-directional microphone arrays. 2.1 Re-annotation procedure Following an initial analysis of the annotations, it was found that a number of issues existed relating to the original cough event annotations in the AMI corpus including. These include incorrec… view at source ↗

**Figure 2.** Figure 2: Examples of inaccurate annotations. Dashed black lines in the figures are the original annotations [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: New annotation criteria, red “x” marks the onset/start of the cough, and green “o” marks the end of the cough event. 4.1 References References [1] M. El Ayadi, M. S. Kamel, and F. Karray, “Survey on speech emotion recognition: Features, classification schemes, and databases,” Pattern Recognition, vol. 44, no. 3, pp. 572–587, 2011. [2] S. M. Schappert and C. W. Burt, “Ambulatory care visits to physician o… view at source ↗

**Figure 4.** Figure 4: GUI created to carry out the re-annotation procedure of cough events in the AMI corpus. 5 [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

read the original abstract

Cough sounds act as an important indicator of an individual's physical health, often used by medical professionals in diagnosing a patient's ailments. In recent years progress has been made in the area of automatically detecting cough events and, in certain cases, automatically identifying the ailment associated with a particular cough sound. Ethical and sensitivity issues associated with audio recordings of coughs makes it more difficult for this data to be made publicly available. However, without the public availability of a reliable database of cough sounds, developments in the area of audio event detection are likely to be hampered. The purpose of this paper is to spread awareness of a database containing a large amount of naturally occurring cough sounds that can be used for the implementation, evaluation, and comparison of new machine learning algorithms that allow for audio event detection associated with cough sounds. Using a purpose built GUI designed in MATLAB, the re-annotation procedure followed a reusable methodology that allowed for quick and efficient importing and marking of audio signals, resulting in a re-annotated version of the Augmented Multi-party Interaction (AMI) corpus' cough location annotations, with 1369 individual cough events. All cough annotations and the re-annotation tool are made available for download and public use.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

This is a straightforward data release that re-annotates coughs in AMI and makes the labels plus a MATLAB tool public, but it is incremental and lacks any validation of the new labels.

read the letter

The paper's core contribution is re-annotating cough locations in the AMI meeting recordings. They built a MATLAB GUI, went through the audio, and produced 1369 labeled cough events, then released both the annotations and the tool for others to use. That directly addresses the scarcity of public cough audio for health-related audio event detection work, which is a real practical constraint in the area. The description of their reusable annotation process is clear enough that someone could replicate the effort on other data if they wanted to. Credit for shipping the actual files and code rather than just describing the idea. The central claim holds up on its own terms: they did the re-annotation and made the output available. No equations or modeling claims to check, so the soundness is mostly about whether the resource exists as stated, which it does. The soft spots are predictable for this kind of paper. There are no numbers on annotation consistency, no inter-annotator checks, no expert medical review, and no side-by-side comparison with the original AMI labels to show what changed or improved. The assumption that the new labels are better is left untested. Scope is narrow by design; this is not advancing detection algorithms or answering a scientific question about coughs. This paper is for researchers who need more labeled cough data to train or benchmark models and are willing to work with meeting-room audio rather than clinical recordings. A methods or applications reader focused on audio ML for health would get some value from the dataset itself. It is honest work with no internal contradictions, so it deserves a serious referee if the venue considers data resources. I would send it out rather than desk reject.

Referee Report

1 major / 2 minor

Summary. The manuscript describes a re-annotation of cough events in the Augmented Multi-party Interaction (AMI) corpus. Using a purpose-built MATLAB GUI, the authors followed a reusable procedure to mark cough locations in the audio signals, producing annotations for 1369 individual cough events. The resulting annotations and the re-annotation tool are released publicly to support development of audio event detection algorithms for cough sounds.

Significance. If the annotations prove accurate, the work is significant for providing one of the larger publicly available collections of naturally occurring cough events. This directly addresses the scarcity of such datasets caused by ethical and privacy constraints, enabling reproducible training, evaluation, and comparison of machine-learning methods for cough detection in health-monitoring applications.

major comments (1)

[Abstract] Abstract: the claim that the release constitutes a 'reliable database' of cough sounds is not supported by any quantitative checks on annotation consistency, inter-annotator agreement, error rates, or agreement with the original AMI labels; the methodology is described only at a high level without such validation.

minor comments (2)

[Abstract] The abstract states that 1369 events were identified but does not report how this count was obtained, whether it differs from the original AMI annotations, or any summary statistics on the re-annotated set (e.g., duration distribution or per-meeting breakdown).
The manuscript would benefit from a short section or paragraph explicitly comparing the new annotations to the original AMI cough locations to quantify the extent of the re-annotation effort.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the review and the recommendation for minor revision. We address the single major comment below.

read point-by-point responses

Referee: [Abstract] Abstract: the claim that the release constitutes a 'reliable database' of cough sounds is not supported by any quantitative checks on annotation consistency, inter-annotator agreement, error rates, or agreement with the original AMI labels; the methodology is described only at a high level without such validation.

Authors: We agree that the abstract's phrasing of a 'reliable database' is not supported by quantitative validation such as inter-annotator agreement, error rates, or comparison to the original AMI labels. The re-annotation was performed by following a consistent single-annotator procedure with the MATLAB GUI, but no such metrics were computed or reported. We will revise the abstract to remove the word 'reliable' and instead describe the release as a publicly available re-annotation of 1369 cough events produced via the documented GUI-based procedure. The methodology section already provides the procedural details at the level implemented; we do not claim additional validation beyond that. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper describes a manual re-annotation of cough events in the AMI corpus using a purpose-built MATLAB GUI, resulting in 1369 annotated events with public release of the annotations and tool. No equations, derivations, fitted parameters, predictions, or self-citation chains appear in the abstract or described content. The central claim is production and release of the re-annotated data, which is satisfied directly by the reported procedure without any reduction to inputs by construction or load-bearing self-references. This is a data-release paper with no mathematical or modeling elements that could exhibit the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

No free parameters or invented entities. The work rests on the domain assumption that human annotation produces reliable cough labels.

axioms (1)

domain assumption Manual annotation by the authors accurately identifies cough events in the audio
The re-annotation count and utility depend on the correctness of the human labeling process without reported inter-annotator agreement or external validation.

pith-pipeline@v0.9.0 · 5743 in / 1125 out tokens · 30663 ms · 2026-05-25T14:14:28.821269+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

5 extracted references · 5 canonical work pages

[1]

Survey on speech emot ion recog- nition: Features, classiﬁcation schemes, and databases,

M. El Ayadi, M. S. Kamel, and F. Karray, “Survey on speech emot ion recog- nition: Features, classiﬁcation schemes, and databases,” Pattern Recogni- tion, vol. 44, no. 3, pp. 572–587, 2011

work page 2011
[2]

Ambulatory care visits to phys ician of- ﬁces, hospital outpatient departments, and emergency depart ments: United states, 2001-02

S. M. Schappert and C. W. Burt, “Ambulatory care visits to phys ician of- ﬁces, hospital outpatient departments, and emergency depart ments: United states, 2001-02.” Vital and Health Statistics. Series 13, Data from the Na- tional Health Survey , no. 159, pp. 1–66, 2006

work page 2001
[3]

Classiﬁcation of human cough sig- nals using spectro-temporal gabor ﬁlterbank features,

J. Schr¨ oder, J. Anemiiller, and S. Goetze, “Classiﬁcation of human cough sig- nals using spectro-temporal gabor ﬁlterbank features,” in 2016 IEEE Inter- national Conference on Acoustics, Speech and Signal Proces sing (ICASSP) , March 2016, pp. 6455–6459

work page 2016
[4]

Unleashing the killer corpus: experiences in creatin g the multi-everything ami meeting corpus,

J. Carletta, “Unleashing the killer corpus: experiences in creatin g the multi-everything ami meeting corpus,” Language Resources and Evaluation , vol. 41, no. 2, pp. 181–190, 2007

work page 2007
[5]

paulleamy/ami-cough-annotations,

P. Leamy, “paulleamy/ami-cough-annotations,” Mar 2019. [Online ]. Available: https://github.com/paulleamy/AMI-Cough-Annotations .git 4 0 2000 4000 6000 8000−1 −0.5 0 0.5 EN2001a.Lapel−2 Sample Amplitude Figure 4: GUI created to carry out the re-annotation procedure of coug h events in the AMI corpus. 5

work page 2019

[1] [1]

Survey on speech emot ion recog- nition: Features, classiﬁcation schemes, and databases,

M. El Ayadi, M. S. Kamel, and F. Karray, “Survey on speech emot ion recog- nition: Features, classiﬁcation schemes, and databases,” Pattern Recogni- tion, vol. 44, no. 3, pp. 572–587, 2011

work page 2011

[2] [2]

Ambulatory care visits to phys ician of- ﬁces, hospital outpatient departments, and emergency depart ments: United states, 2001-02

S. M. Schappert and C. W. Burt, “Ambulatory care visits to phys ician of- ﬁces, hospital outpatient departments, and emergency depart ments: United states, 2001-02.” Vital and Health Statistics. Series 13, Data from the Na- tional Health Survey , no. 159, pp. 1–66, 2006

work page 2001

[3] [3]

Classiﬁcation of human cough sig- nals using spectro-temporal gabor ﬁlterbank features,

J. Schr¨ oder, J. Anemiiller, and S. Goetze, “Classiﬁcation of human cough sig- nals using spectro-temporal gabor ﬁlterbank features,” in 2016 IEEE Inter- national Conference on Acoustics, Speech and Signal Proces sing (ICASSP) , March 2016, pp. 6455–6459

work page 2016

[4] [4]

Unleashing the killer corpus: experiences in creatin g the multi-everything ami meeting corpus,

J. Carletta, “Unleashing the killer corpus: experiences in creatin g the multi-everything ami meeting corpus,” Language Resources and Evaluation , vol. 41, no. 2, pp. 181–190, 2007

work page 2007

[5] [5]

paulleamy/ami-cough-annotations,

P. Leamy, “paulleamy/ami-cough-annotations,” Mar 2019. [Online ]. Available: https://github.com/paulleamy/AMI-Cough-Annotations .git 4 0 2000 4000 6000 8000−1 −0.5 0 0.5 EN2001a.Lapel−2 Sample Amplitude Figure 4: GUI created to carry out the re-annotation procedure of coug h events in the AMI corpus. 5

work page 2019