A Bayesian Modeling Approach to Situated Design of Personalized Soundscaping Algorithms

Round 1

Reviewer 1 Report

In this paper the authors perform a Bayesian Modeling Approach to Situated Design of Personalized Soundscaping Algorithms. The authors have divided the paper into eight sections, I recommend reviewing the division of information into a smaller number of sections. For example: Introduction, Materials and Methods, Results, Discussion, Conclusions.

 

Section 1 must be improved. It is necessary to add more references, specifically in the first part, to allow the non-expert reader to deepen the topics. Also, before explaining in detail what this study is about, it is necessary to properly introduce the reader to the subject. you could propose works that have already dealt with the separation of sources, and so on.

Section 2 must be improved. Section 2, 3, and 4 can be joined in a single Section named Materials and Methods. Each section can be added as a subsection.

Section 3 must be improved. Section 2, 3, and 4 can be joined in a single Section named Materials and Methods. Each section can be added as a subsection.

Section 4 must be improved. Section 2, 3, and 4 can be joined in a single Section named Materials and Methods. Each section can be added as a subsection.

Section 5 must be improved. Enhance graphs showing rating metrics. Add labels to subplots. Describe in detail the metrics used to evaluate the performance of your method (SNR, PESQ, and STOI).

Section 6 must be improved. Move this section in the introductory section of the paper. There is no point in describing the work of other authors after showing their results. This information should be proposed at the beginning of the paper in order to show how other authors have approached the problem.

Section 8 must be improved. Paragraphs are missing where the possible practical applications of the results of this study are reported. What these results can serve the people, it is necessary to insert possible uses of this study that justify their publication.

 

114-115)Introduce adequately the three main factors: soundscaping, source mixing and source modeling.

127) Do not use abbreviation such as i.e. I have seen that you often use this abbreviation, so I will not repeat this advice again, it also applies to the other occurrences.

233) the factorizable probability density function formula is too long in this line, move this in the next line

246)” Furthermore 6 shows” Maybe, Furthermore Figure 6 shows

457) Add a label for each subplot in Figure 9. Add this label in the Figure caption.

457) Add a label for each subplot in Figure 10. Add this label in the Figure caption.

444-445) Describe in detail the metrics used to evaluate the performance of your method (SNR, PESQ, and STOI).

Author Response

Dear reviewer,

Please find enclosed our response in the attached pdf document.

Yours sincerely,

Bart van Erp

Author Response File: Author Response.pdf

Reviewer 2 Report

The work proposes a novel framework for adapting the enhancement process in hearing aids in order to improved the quality of the perceived acoustic environment (therefore called "personalized soundscaping"). The authors clearly introduce the problem and  formulate the design method following a Bayesian approach. The proposed solution is sounding and my minor concerns are related to the experimental part where only a limited analysis is presented. As far as I understand, only one noise type is considered (clapping):  simulations with other noises (even in case of a mixture of noises) would strengthen the conclusions of the paper, demonstrating the generality of the approach and its robustness against a variety of background noises (e.g. discussing the case of multiple and overlapping noises). Considering also that the improvements in terms of PESQ/STOI seem small for some SNRs, maybe some listening tests can validate the proposed technique. Another point that deserves a more thorough discussion is related to computational analysis, since it is not clear whether the algorithms can be realistically run on the targeted hearing devices.

Overall the paper is well written and convincing; some additional experiments can provide evidence to the effectiveness of the framework in very general acoustic conditions.

Author Response

Dear reviewer,

Please find enclosed our response in the attached pdf document.

Yours sincerely,

Bart van Erp

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have only partially incorporated the suggestions provided by the reviewer, the form and content of the paper seem substantially unchanged compared to the previous version, just analyze the changes made to the document. There are sections that need to be reviewed. The authors have left out some suggestions considering some sections sufficiently clear, evidently only for them. An article is not intended only for experts in the subject but must also be of interest to other readers who want to approach the methodology, perhaps to apply it in other sectors. Therefore more attention is needed to these details.
My request to move section 6 to the introductory section of the document is due precisely to these considerations. If the reader does not know perfectly how the problem was tackled by other authors, can he or she appreciate the work of the authors? Therefore it makes no sense to describe the work of other authors after showing their results. This information should be proposed at the beginning of the article to show how other authors have approached the problem.

Author Response

Dear reviewer,

Please find enclosed our response in the attached pdf document.

Yours sincerely,

Bart van Erp

Author Response File: Author Response.pdf

Round 3

Reviewer 1 Report

The authors addressed all the reviewer's comments with sufficient attention and modified the paper consistently with the suggestions provided. The new version of the paper has improved significantly both in the presentation that is now much more accessible even by a reader not expert in the sector, and in the contents that now appear much more incisive.

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

This paper proposed a probabilistic framework on adaptive denoising for hearing aids. The comments are as follows.

 

* The novelty of the proposed method is not supported by the literature. The background section includes only 8 citations, and these are in only one sentence at L32. The authors have to pay respect to the previous researches and have to introduce the problem clearly with literature. The same goes for section 2.

Also, none of the literature is about hearing aid signal processing. The readers cannot understand the context of the hearing aid's research, and what exactly is the key contribution of this research. The cocktail-party effect is a very general problem and this effect itself is not specifying the problem.

For example, the history of the signal processing for hearing aid is reviewed occasionally, e.g. [1], [2], or [3]. The research should be placed in the context of the research field.

[1] Chung K. Challenges and recent developments in hearing aids. Part I. Speech understanding in noise, microphone technologies and noise reduction algorithms. Trends Amplif. 2004;8(3):83-124. doi: 10.1177/108471380400800302. PMID: 15678225; PMCID: PMC4111442.

[2] Launer S., Zakis J.A., Moore B.C.J. (2016) Hearing Aid Signal Processing. In: Popelka G., Moore B., Fay R., Popper A. (eds) Hearing Aids. Springer Handbook of Auditory Research, vol 56. Springer, Cham. https://doi.org/10.1007/978-3-319-33036-5_4

[3] Magudilu Srishyla Kumar Lakshmi, Ayasakanta Rout & Cynthia R. O’Donoghue (2019) A systematic review and meta-analysis of digital noise reduction hearing aids in adults, Disability and Rehabilitation: Assistive Technology, DOI: 10.1080/17483107.2019.1642394

 

* The experimental setup is not clear, thus cannot understand the results. For example, the problem statement introduced a need for user control in Fig.3, but never mentioned the subjects in this experiment. It seems the experiment was only conducted with numerical simulations. In this case, w^{1:K} cannot be estimated and should be given. However, the weights are not mentioned in the experimental setup. The same goes for other variables.

 

* The results are only shown for the proposed method. This makes it impossible to evaluate the performance against other approaches. Is the proposed method improved any criterion from previously-proposed methods, and how much is it? These basic questions were not answered in the paper.

Reviewer 2 Report

In this paper the authors perform a Bayesian Modeling Approach to Situated Design of Personalized Soundscaping Algorithms.

The Introduction must be improved. It is necessary to add more references, specifically in the first part, to allow the non-expert reader to deepen the topics. Also, before explaining in detail what this study is about, it is necessary to properly introduce the reader to the subject. you could propose works that have already dealt with the separation of sources, and so on.

Section 2 must be improved. The basic arguments have not been adequately introduced. The equations seem unnecessarily abstract without a proper introduction. Methodologies are also mentioned which are not adequately introduced such as Kalman filtering.

Section 3 it is unnecessarily complex. Many equations to explain the final model. Authors must make an effort to make it simpler, so that it is understandable even to an inexperienced reader.

Section 4 the arguments are introduced adequately even if there is no lack of methodologies only cited without an adequate explanation.

In Section 6 the audio files used for model validation must be explained in more detail. In addition, detailed information on the hardware and software requirements used for the simulation must be entered. Finally, it is necessary to clarify which software applications have been used: ForneyLab, etc. You could use a diagram in which starting from the signal you indicate the software used for processing the signal up to the final result. This will make it clear to the reader how to reproduce the experiment.

In Section 7 authors should summarize their contributions to this study. Since the platform to carry out the processing was ForneyLab, it is necessary to specify how they used this platform. Did they participate in the development of the platform or did they set the platform for this specific application?

 

 

9) remove i.e. abbreviations

12) use the complete definition of ViSQOL acronym or remove this indication

28)” situated soundscaping” Explain or add references to deep the topic

31-33) Briefly list the topics of the papers, or use a simple indication [1-8]

25-37) Add references to support this statement

39)” situated soundscaping” Introduce the topic

41) dynamic latent variable model - Introduce the topic or add references

49) do not use i.e. abbreviations. I've seen you use this abbreviation often, so I won't mention it in the rest of the paper.

84) to be to be statistically independent - ?

81-87) The basic arguments have not been adequately introduced. On line 79-81 you say: Here we shortly discuss these stages in more detail. Add references to these topics.

87)At the end of equation (2) remove a period. You also need to specify all terms that appear in the equation

92) Kalman filtering - Introduce the topic or add references

94) smartphone app - this is the first time we talk about smartphone apps. Are you implementing an application for a smartphone?

105-108) Specifies that this model will be adopted in all three phases: source modeling, source separation and soundscaping

109) i.i.d. noise Use the complete definition

110) diagonal precision matrix – introduce the topic or add references

122-123) Start the section by proposing studies that use Gaussian methodology in real life problems. For example: “Gaussian regression for recycled polyethylene”.

133) Gaussian random walk – Introduce this topic

140) N > 2M Explain what is N and what is N. Add references to support this statement

160-161) I think it is more appropriate to first introduce factor graphs and then show Figure 4.

161) A period is missing

187) marginal distribution -  Introduce the topic

201-203) Add references to these methodologies

266-268) Better specify the data used for training and testing the model

270-273) Has your model been somehow trained in the ForneyLab environment?

275-283) Better specify which signals the generative model has been trained on. Specify type, number, duration etc.

287) Figure 9 is very illustrative. The three graphs must be labeled with a, b and c so that they can be referred to clearly and uniquely in the caption.

352-378) The possible practical applications of the results of this study and the possible future goals of this work, must be moved in the conclusion section.