The technology of audio stream segmentation and speech recognition created by the MDG group of companies was recognized as the best at the international chime Speech Separation and Recognition Challenge (CHiME-6). This was announced on Thursday, May 7, by the developer’s press service.
“At CHiME-5, the contestants solved the so called cocktail party problem-recognizing the spontaneous speech of several speakers in conditions of partial overlap of speech and noise, that is, in a typical situation of communication at a party. This unit is required to operate the segmented (already allocated) speech. The novelty and feature of CHiME-6 was that for the first time in history, the contestants were asked to solve a similar problem, but work with non-segmented speech, while-with speech overlap of up to 20%, ” the press release says.
Entries for the competition were made at 20 dinners in real homes, where people freely communicated, joked, laughed, cooked, ate, and washed dishes.
The organizers set a goal for participants to create a recognition system that will “listen” to recordings and give a complete transcript with the least number of errors. As a result, the winner was the technology created by MDG specialists.
“High quality speech recognition of different speakers, while interrupted by noise, allows you to bring services from the category of innovative to everyday use, improving business and simplifying our lives,” said Dmitry Dyrmovsky, CEO of the MDT group of companies.
In early January, Russian President Vladimir Putin said in his annual address to the Federal Assembly that the country is able to achieve a breakthrough in the development of artificial intelligence.
In November of last year the savings Bank has created the most powerful Russian supercomputer. This model helps speed up the development of services and processes based on artificial intelligence.