Evaluating the Variance in the Threshold of Authenticity for Synthetic Speech Verification — Volunteers and Related Parties


This test was designed to test an individuals effectiveness in picking out synthetic media when the threshold of authenticity has been varied. This link has been created for The people who very kindly volunteered their voice for this test, and those who know them. So as not to influence the results in any way, this test will not generate any results but merely serves as a demonstration of the real test and allows those who can’t participate in the real test to still experience it.


The “Threshold of Authenticity” is defined here, as the level at which it becomes clear that a file is either real or synthetic. The theory in this test is that as additional factors affect the speech, does this make synthetic audio harder to detect and conversely does it add doubt to whether or not the real audio files are indeed real.


When you receive a phone call, you do not expect it to be studio-quality and likewise when hearing a person speak through a wall, the sound becomes muffled. Therefore it is theorized that if a piece of synthetic speech is manipulated in such a way as to replicate these scenarios, an individual will be less likely to think it is synthetic.


Each question features 3x audio files to choose from. The audio files in this test are a mix of real speech and synthetic speech. Each question is designed to propose a different scenario which alters the threshold of authenticity of the files in an attempt to make the participant question which files are real and which are synthetic.

Each question features 3x audio files to choose from, which will either contain real speech or synthetic speech. When answering the question, listen to each audio file and use the check boxes below to select which files you believe contain synthetic speech. The question will contain between 0-3 synthetic files. Each question is designed to propose a different scenario which alters the threshold of authenticity of the files in an attempt to make the participant question which files are real and which are synthetic.


If you have any questions or you would like your results after taking this, please email me at j.rodgers@ljmu.ac.uk


Q1: Are any of these audio files a deepfake?
Select all that apply.
Q2: Are any of these audio files a deepfake?
Select all that apply.
Q3: Are any of these audio files a deepfake?
Select all that apply.
Q4: Are any of these audio files a deepfake?
Select all that apply.
Q5: Are any of these audio files a deepfake?
Select all that apply.
Q6: Are any of these audio files a deepfake?
Select all that apply.
Q7: Are any of these audio files a deepfake?
Select all that apply.
Q8: Are any of these audio files a deepfake?
Select all that apply.
Q9: Are any of these audio files a deepfake?
Select all that apply.
Q10: Are any of these audio files a deepfake?
Select all that apply.