Today,
SWEALINK proposes
VOX PROTECT.
EXPLANATIONS ABOUT VOX PROTECT
After several years of development and research,
Its researchers and sound engineers have developed an innovative and non-reversible solution.
First, the voice to be faked undergoes two linear transformations resulting from new technologies MOD A and MOD B.
The software cannot reverse these two transformations.
-The effect obtained cannot be reversed by a simple pitch, getting the first level of protection (low level).
Then, these two transformations have three levels of random computation, VAR A, VAR B and VAR A+B, making it impossible to reverse the calculation, even by the software itself. The simultaneous use of these 3 random effects allows for reaching the second level of protection (high level of protection).
At this level, it is essential to know that the algorithm never repeats the same random calculation twice, making it impossible to “search” for variations.
The plug-in and the operator do not keep any trace of random calculations operation,
protecting all the “sources”, that is to say :
-the witness,
-the user of the software,
-the software itself,
thus, making the process irreversible.
In certain sensitive cases, journalists use an actor’s voice to replace the voice of the witness to protect his identity.
The software will be helpful to protect the actor’s identity as well.
As proof of the effectiveness of VOX PROTECT,
the analysis of the sonogram, the spectral representation of the voice’s harmonics, before and after treatment, clearly shows the effectiveness of the VOX PROTECT software.
Let’s take the example of a witness’s voice treated by a pitch:
SONOGRAM OF THE ORIGINAL VOICE WITHOUT ANY ADDITIONAL PROCESSING
SONOGRAM OF THE VOICE TRANSFORMED BY A PITCH AT -3
We can see that the graphical representation of the harmonics has not undergone any significant change, making voice recognition possible.
If we apply a double pitch to the original voice, it is the same case.
Now let's take the example of the same original file processed by VOX PROTECT in high-level protection:
SONOGRAM OF THE ORIGINAL VOICE WITHOUT ANY ADDITIONAL PROCESSING
SONOGRAM OF THE VOICE PROCESSED BY VOX PROTECT IN HIGH LEVEL OF PROTECTION
The VOX PROTECT effect acts on the signal’s harmonics, making speech recognition impossible.
STUDY ON THE TWO LEVELS OF PROTECTION proposed in the software preferences:
After studying the use cases, we have defined two levels of protection:
The first level of protection “Low level of protection” corresponds to weak security.
This “low level” protection is more reliable than a common effect such as pitch.
Indeed, the “low level” effect found cannot be reversed by the most common “audio effects” on the market.
On the other hand, it remains reversible and can be analyzed by more “specialized” services.
We advise you to use this level of protection for witnesses, for example, who testify on facts without significant risk of recognition or for the actor’s voice replacing the voice of the witness.
The second level of protection, called “High level of protection” corresponds to a strong level of protection that protects, for example, a witness who denounces a drug trafficking or a terrorism case.
This level will protect in all situations. The downside is that your voice will become less understandable and require subtitles in some cases.
We have given some examples of preferences, classified by voice type, in the “presets” folder for information purposes only.
PRESETS
The “presets” are organized in two categories:
LOW PROTECTION/REVERSIBLE
)
HIGH PROTECTION/NOT REVERSIBLE
The low protection corresponds to light protection, which protects only from a reverse operation of a pitch, for example. This level does not protect against an attempt by voice recognition.
We offer 5 possible presets for each voice category.
For example, for a high female voice, we have: Low woman high pitched no variation numbered from 1 to 5
)
For the LOW PROTECTION/REVERSIBLE level, we have decided to propose alternative low protection preset with a slight variation, thus reinforcing the level for each voice category
For example, for a high-pitched female voice, we have: Low woman high, pitched variation numbered from 1 to 5.
)
For the LOW PROTECTION/REVERSIBLE level, we have decided to propose alternative low protection preset with a slight variation, thus reinforcing the level for each voice category
For example, for a high-pitched female voice, we have: Low woman high, pitched variation numbered from 1 to 5.
)
This level makes the signal less readable, so it will be up to the user to decide if he wants to lower specific criteria depending on the characteristics of the voice to be faked.
Important recommendation
The voice is a complex audio signal and is specific to each case.
We inform that depending on the characteristics of the voice to be faked, it will be the user who will decide the final levels of protection, that is to say, the final levels for each criterion and this for each witness to be protected.
We recommend working in an audio suite or offline when using VOX PROTECT.
Working in insert mode is limited in the number of tracks used simultaneously. For example, on Pro Tools, the maximum use of two tracks in the insert is recommended.
For tips on using VOX PROTECT, we recommend you watch the video and follow our FAQ.
Get
Vox Protect
The only software dedicated to witnesses’ voice protection offers a unique and maximum level of anonymization while keeping the original emotional intonations.