Scientists from Tomsk University of Control Systems and Radioelectronics have created a mathematical model that recognizes speech and outperforms its peers with specific indicators.
The innovators point out that this model could become the basis for a comprehensive system of personal identification, as well as facilitate the control of “smart” technology with the help of voice.
According to experts, the method of personal identification is being developed through speech recognition using digital security technology used to control equipment and security systems in remote banking services.
Scientists from Tomsk University of Control Systems and Radioelectronics have created a new mathematical model of speech recognition, which surpasses its peers in accuracy and allows the correct identification of the speaker.
Anton Konev, Deputy Director of the National Institute of Communications Competence Center “Reliable Interaction Technologies,” says: “The mathematical apparatus differs from its counterparts in its great similarity to the human inner ear when recognizing critical speech frequencies.
The potential mechanisms underlying the current speech recognition systems in this device are only auxiliary.”
Experts point out that the error rate in the new device’s recognition of the basic tone of speech is four times less than the best analogs in the world, and this allows for better recognition of normal speech and even singing. It also allows communicating with smart devices not through commands, but directly, as is the case with the interlocutor.
Experts assure that the new system will allow verifying the identity of the interlocutor during the communication session, which will completely eliminate some types of fraud.