Alzheimer’s Dementia Speech (Audio vs. Text): Multi-Modal Machine Learning at High vs. Low Resolution

Round 1

Reviewer 1 Report

This study investigates the effectiveness of different machine learning techniques for automatically detecting Alzheimer's dementia. These techniques include audio signal based as well as text based ones. The study compares and discusses the results of these approaches. While the article certainly has a potential, some issues need to be addressed before it can be considered ready for publication.

The biggest weakness of the article is that the authors do not disclose details about the number of participants or the recording techniques (at least not in an adequate and clear form). Some insight is provided on lines 241-251, but that's very inelegant given the overall organization of the text. This is a key aspect and should be communicated better.

The authors have made good use of the rich portfolio of available tools and ready-made solutions. In this regard the article is excellent. However, this sometimes has a negative impact on the readability of the text. Thorough reflection on the logical structure of the manuscript would be advisable. For example:

- It would be good to clearly define the frame-level and the file-level scopes somewhere in the introduction or at the beginning of Methods.

- The abstract is vaguely written and poorly reflects the strengths of the text. For instance, “Sixteen features, including four feature extraction methods not previously applied in such contexts…” Which ones?

- There are many incoherencies. Such as that the abbreviation CN first appears on line 20, but is explained on line 79.

I have doubts about the image-based model based on energy-time plots. Since we don't know the details of the recording technique and the data are normalized, could it be, for example, that the model responds to different levels of signal-to-noise ratio, but that this is an artifact of the normalization?

Minor errors and typos:

- Inline equations really should not be written as plain text (line 130, for instance). The reader should have the variables clearly distinguished from the rest of the text, abbreviations, etc.

- line 317: It would be good to use a better formulation than “signal:noise”.

Author Response

Thank you for the kind feedback. We have incorporated your suggestions in the updated manuscript. Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper " Alzheimer’s Dementia Speech (audio vs text): Multi-model Machine Learning at High vs Low Resolution" falls within the scope of Applied Sciences Journal and shows some technical relevance.

In this paper, the authors investigate a multi-modal approach (audio and text) for the automatic detection of Alzheimer’s Dementia determined from recordings of spontaneous speech.  

The material is publishable but requires improvement. In this sense, there are some remarks and or suggestions on the attached paper that should be addressed before publishing.

Suggestion 01

The abstract does not entirely fulfil the function of a summary. Therefore, the abstract should be reviewed to give, in addition to the numerical key results, research gap and aim, and the methodology applied.

Suggestion 02

It is advisable not to use the grouping of references as this is detrimental to the accuracy of the contextualisation of the topic in the scientific literature.

Suggestion 03

The structure of the paper is clear and adequate, however some key sections, such as the Introduction, should be revised. In the first part of the introduction some excessively generalistic and redundant expressions are used. Some statements and data are included without being linked to the necessary literature references.

Suggestion 04

The importance of the problem and the pertinence of the study must be clearly stated. Reference’s structure should be more clearly defined, including specific results of the most significant studies. In short, the Introduction does not include a comprehensive literature review.

Suggestion 05

The limitations of the study should also be included in the Introduction Section. Given the subject matter, a good definition of the limitations of the study is considered particularly important. This issue will be discussed below.

Suggestion 06

Novelty unclear: What is the original contribution of the study? The paper is not very enlightening on the subject. The novelty should be made as clear as possible.

Suggestion 07

The model used in the calculations of the Machine Learning algorithms needs to be described in more detail. The number of samples taken as input data and for model accuracy checks must also be specified.

Suggestion 08

The Discussion and Conclusions Sections are formalised with greater scientific rigour. However, the reproducibility of the study cannot be guaranteed based on the data and descriptions provided in the manuscript in Section 2. Experimental Methodology. It is therefore strongly advised to review this section.

Author Response

Thank you for the kind feedback. We have incorporated your suggestions in the updated manuscript. Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper analyzes a very interesting and current topic, which can be transferred to several areas.

The introduction is comprehensive and measured.

In the methodology section, describe in more detail the data collection and sampling method.

I could not find this reference " Clarke, C.J.; Melechovsky, J.; Lin, C.M.Y.; Priyadarshinee, P.; Balamurali, B.; Chen, J.M.; Kapoor, S.; Aharonov, O. ADDRESSING MULTI-MODAL MULTI-MODEL MULTI-FEATURE CUES IN ALZHEIMER’S DEMENTIA.". You state in the paper that it agrees with your results. Also, data is missing for this reference.

Can you compare the results with some other references in the discussion?

In conclusion, add directions for future research, limitations, and practical application of your results.

Author Response

Thank you for your valuable feedback. We have now incorporated your comments in our manuscript. Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

My concerns were addressed properly and the article is ready to be accepted from my point of view.

Reviewer 2 Report

No additional corrections to the manuscript are considered necessary.