Rejection for potential voice biometric vulnerabilities

Christine Hanson 1 Reputation point
2021-10-15T21:09:07.68+00:00

I could not find any documentation regarding the following two cases for voice recognition in MS Azure:

1- if a recording of a previous enrollment is used for verification, is it rejected?

2- if a single recording of a previous verification is re-used multiple times, is it rejected?

Both cases are security vulnerabilities and how does Azure Speech address them?

Azure AI Speech
Azure AI Speech
An Azure service that integrates speech processing into apps and services.
1,337 questions
{count} votes

1 answer

Sort by: Most helpful
  1. romungi-MSFT 41,736 Reputation points Microsoft Employee
    2021-10-18T10:09:19.163+00:00

    @Christine Hanson Are you planning to use the Speaker recognition preview feature for authentication purposes in your application?
    According to the documentation and guidance of this feature it is not intended to use the speaker recognition for these purposes.

    The APIs are not intended to determine whether the audio is from a live person or an imitation/recording of an enrolled speaker.

    With respect to the above questions, for a text dependent verification a specific passphrase should be used during enrollment and the same is used for verification. If you use the same recording then I think the verification will be successful.

    For the second question, the same scenario should apply for text dependent verification. For text independent verification then atleast 20 seconds of total audio is required and a recording could be used successfully to verify the person.

    Since the use-case for speaker verification is mostly based on improving remote meeting productivity by identifying users by voice profile, I think using the feature for authentication purpose is not appropriate.

    If an answer is helpful, please click on 130616-image.png or upvote 130671-image.png which might help other community members reading this thread.

    0 comments No comments