NRNU MEPhI: Speech signature to protect documents from forgeries and falsifications

The main articles are:

• Speech Recognition (Technology, Market)
• Biometric identification technologies
• Cryptography

2023: Speech Signature Presentation

Scientists of NRNU MEPhI on August 16, 2023 presented their development under the name "Speech signature to protect documents from fakes and falsifications."

Classic and electronic signatures are known to be vulnerable. An attacker can later edit a document signed with a pen by assigning, for example, the numbers he needs, since in this case there is no connection between the protected text, the author and the signature. An electronic signature already gives a connection between protection and text, but does not give a connection between the author and protection, since an attacker can use someone's token to sign a document. The voice signature protects the document in full.

To create a speech signature linking the author, text and protection together, a person speaks with his voice the main points of the document that he wants to protect - the terms, amounts, obligations of the parties, even the entire document, if necessary. Then the "voice signature" is converted into a picture, inserted at the end of the document and recognized using the software.

biometric data Other speakers can also be included in the speech signature, measured "smart" watches using -, pulse pressure, temperature. This is necessary so that later it is possible to assess the psycho-emotional state of a person at the time of signing the document and exclude the influence of external factors - pressure, threats, etc.

"It is impossible to fake a voice signature - for this you would have to speak in the voice of another person, but one similarity will not be enough, and the voice generated by special systems is perfectly recognized by technology by certain tags. This is like in the case of generated paintings - with the help of special programs, you can always establish the original picture in front of us or created by a neural network, - explains Alexander Alyushin, researcher at the Scientific and Educational Center "Security of Intelligent Cyber ​ ​ Physical Systems" (REC "BIX") NRNU MEPhI. "In general, we have been engaged in speech technologies for 10 years, we have deduced a certain pattern in the formation of a speech signal, based on which we developed these algorithms."