All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
The authors have significantly revised the article in light of the 2nd revision comments.
The authors have not correctly and thoroughly revised the article based on previous comments. They should,
1. highlight their contributions with respect to Melissa N. Stolar; Margaret Lech 2017 and 2020
2. Add details of pipeline for extracting spectrogram feature
3. Address concerns related to validity of the findings
[# PeerJ Staff Note: Please ensure that all review and editorial comments are addressed in a response letter and any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate. It is a common mistake to address reviewer questions in the response letter but not in the revised manuscript. If a reviewer raised a question then your readers will probably have the same question so you should ensure that the manuscript can stand alone without the response letter. Directions on how to prepare a response letter can be found at: https://peerj.com/benefits/academic-rebuttal-letters/ #]
NA
1. The authors now cited Melissa N. Stolar; Margaret Lech 2017 and 2020 which follow a similar approach to the proposed method however the similarities and differences in the proposed approach with these references in particular and other approaches, in general, must be highlighted.
.
2. The authors have explained in the response letter their rationale and strategy for feature selection from speech. However, it will be helpful for replication, etc. to add the implementation details in the manuscript itself about the pipeline for extracting spectrogram features and then selecting the most suitable.
As a response to the justification for classification metrics used in the results section e.g. the confusion matrix, the authors have referred to Table 3 which compares the nature of speech datasets. The suggestion here was to explain the reasons for using certain classification metrics/confusion matrices, what they indicate and what are the implications of the obtained results in real-world scenarios.
NA
The authors should revise the manuscript in light of both reviewers comments considering major revisions.
[# PeerJ Staff Note: Please ensure that all review and editorial comments are addressed in a response letter and any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate. It is a common mistake to address reviewer questions in the response letter but not in the revised manuscript. If a reviewer raised a question then your readers will probably have the same question so you should ensure that the manuscript can stand alone without the response letter. Directions on how to prepare a response letter can be found at: https://peerj.com/benefits/academic-rebuttal-letters/ #]
[# PeerJ Staff Note: The review process has identified that the English language must be improved. PeerJ can provide language editing services - please contact us at copyediting@peerj.com for pricing (be sure to provide your manuscript number and title) #]
The paper is very well written however at several places poor sentence structure makes the paper difficult to read and create ambiguities. Professional proof reading is required before publication.
The contribution of the paper is the field is appreciable but the following shortcomings and must be addressed. .
In their proposed approach the authors have used the AlexNet for feature extractions. This network architecture is too old and has been superseded by several newer CNN based feed forward networks. Particularly the use of 11X11 convolutions in the AlexNet has been mostly discontinued and 3x3 kernels are used in most current networks.
The authors have used the transfer learning approach by using a pretrained AlexNet. However, they have not clearly mentioned the dataset used for pretraining of the Alexnet.
The authors do not specified the details of deep learning python libraries (e.g. Tensorflow/Keras or Pytorch) etc. in section 4.2.
In sub-section 5.0.1 the the number of neurons, activation functions and loss functions used to train the MLP classifier have not been specified. Similarly, hyperparameters used for the other methods are also not mentioned.
The findings have been evaluated by using the well known metrics in the field and conclusion is well stated.
Language needs to be improved including the following particular issues:
Line 71: ‘have been founded’ should be ‘have been found’
Line 77-78: 'Therefore' opening the first two consecutive sentences
Line 109: ‘Emotions are categorized into two approaches’ should be ‘Emotions are categorized using two approaches’
Section 2.1 and literature review in general presents various kinds of techniques and their results without a chronological sequence
The method proposed in this paper extracts Alexnet based deep learning features from spectrogram for speech and then uses correlation based feature selection before feeding multiple shallow classifiers for emotion recognition. However the manuscript should address following two main issues:
1st issue:
A similar method is proposed by ‘Melissa N., et al 2017 and 2020’. They treat the speech spectrogram as an RGB image and uses Alexnet for feature extraction to classify emotions without the ‘feature selection step though.
The manuscript has not referred to these papers at all. The authors must highlight how the proposed method is different or better from the mentioned work.
Stolar, Melissa N., et al. "Real time speech emotion recognition using RGB image classification and transfer learning." 2017 11th International Conference on Signal Processing and Communication Systems (ICSPCS). IEEE, 2017.
Lech, Margaret, et al. "Real-time speech emotion recognition using a pre-trained image classification network: Effects of bandwidth reduction and companding." Frontiers in Computer Science 2 (2020): 14.
2nd issue:
The experimental setup for feature selection from speech features is not explained in detail. Both sections 3.4 and 3.5 provide the generic foundation of feature selection techniques without any explanation of implementation in the proposed setup.
The results section presents confusion matrices for different datasets without explaining their significance. The manuscript should justify the classification metrics analyzed in results in terms of their significance e.g. what are the consequences of classifying a particular emotion as another and what the confusion matrices aim to highlight.
Tables 3-6 present accuracies for different classifiers without mentioning ‘accuracy’ in the table tile or any axis
Line 398 in conclusion suggests that authors aim to test the results in cloud and edge computing environments. What is the rationale for such tests?
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.