Pathological voices extension

Hello all.

I’d like to propose extending CommonVoice’s functionality for pathological voices.

I am researching the automatic diagnosis of pathological speech [1]. The variation of pathological voices is much wider than that of healthy voices. Therefore, it is desirable to create a database of pathological voices that is larger than the database of healthy voices. However, at present, the databases of pathological voices are much smaller than the databases of healthy voices, and with a few exceptions [2, 3], they are private ([3] is inspired by CommonVoice). I think this is a barrier to the development of my research and all research on the automatic diagnosis of pathological voices.

I think that CommonVoice is a great project. I think CommonVoice is a great project, and I think that adding components related to pathological voices to this project will help advance the study of pathological voices. Specifically, I think the following should be added:

  1. optional information on voice disorders (e.g., healthy, laryngeal cancer, vocal fold paralysis, dysarthria, Parkinson’s disease)
  2. recording of sustained vowels

Sustained vowels are a common utterance content used in the studies of pathological voices. However, there is a shortage of data on sustained vowels not only in pathological voices but also in healthy voices.

I would appreciate it if you would consider it.

Thank you very much.

[1] https://www.isca-speech.org/archive/Interspeech_2020/pdfs/3228.pdf

[2] http://www.stimmdatenbank.coli.uni-saarland.de/index.php4#target

[3] https://merriekay.com/uncommonvoice

Dear @Shunsuke_Hidaka, thank you for your interest. I think this is unlikely to happen and would be better in a separate project because of privacy concerns. E.g. I doubt Common Voice wants to be storing information about people’s medical conditions as there are a lot of regulations surrounding that. The Common Voice code is free/open-source and there is also a free/open-source android app that has been developed.

However I am not a representative of NVIDIA or Mozilla so I cannot speak for the project in general, I can just give you my personal opinion.

Best regards and good luck with your research,

1 Like

@ftyers Thank you for your reply.

I think this is unlikely to happen and would be better in a separate project because of privacy concerns. E.g. I doubt Common Voice wants to be storing information about people’s medical conditions as there are a lot of regulations surrounding that.

I think it is as you have pointed out. Information about diseases is more personal than information about age, and collecting information about diseases seems far from the purpose of CommonVoice.

Your opinion has helped me to rethink my thoughts. Thank you very much.

1 Like

You might want to collect them under your own terms, and push them to Common Voice instance by Mozilla later (I’m unsure whether it is doable right now or if it requires work, @phire would be able to tell us) by stripping the medical level infos.

It would require you to maintain a fork of Common Voice, obviously.

2 Likes

Thank you. I see that there is a possibility to tie it to Common Voice in that way.

I have come to realize that what I posted is very different from what Common Voice is aiming for, and it is not appropriate to directly combine them.

I am sorry for making such a shallow suggestion before I understand Common Voice’s philosophy. I am currently reading the CommonVoice Discourse to deepen my understanding. I think Common Voice is a great project. So, regardless of pathological speech, I would like to contribute to Common Voice through recording and validation, etc.

1 Like

Thanks for the tag in @lissyx. As @ftyers says, we’re definitely not going to collect medical information from individuals. This is partially for privacy and regulatory compliance reasons, but I would also personally be very concerned about the potential for abuse. As all of the data we collect is released under a CC-0 license, we we have no control over how it gets used once it’s in the public domain, and I could easily see that kind of metadata being used to train models that learn to discriminate as well as diagnose.

The Common Voice platform currently does not accept bulk imports of audio that was created elsewhere, nor are we likely to create the functionality to do so, as it would be difficult to guarantee that all sentences read are CC-0 and all the data was collected in such a way that conforms to Mozilla’s privacy and terms of service in respecting the rights of the data subject. For this specfic example, I also imagine additional guidance would be needed to help contributors learn how to validate recordings with speech disfluencies so that those speakers are not unfairly penalized, and the platform is not currently set up to provide more sophisticated guidance on how to validate. For all of those reasons, a fork of the CV platform that is adapted specifically for those communities probably makes more sense.

3 Likes

Thank you for explaining in detail the concerns that we have because we are public (especially CC-0). I agree with the points you made. You have helped raise my awareness of compliance and risk management. I think I was focusing only on the fact that it would be beneficial to have a large database of pathological voices and was not very concerned about the risks.

@ftyers, @lissyx, @ftyers, you all are right, I now understand that even if we do create a database of pathological voices, it should be separated from Common Voice.

I am very grateful for the sincere comments on my proposal, which is far from the purpose of Common Voice. Thank you very much.

2 Likes