Russian subtitles based on speech recognition technologies appeared in the VKontakte player

The social network “VKontakte” has launched a new function of subtitles in Russian in its video player. They are automatically generated based on proprietary speech recognition technologies, machine learning algorithms, and intelligent noise reduction technologies used on the video calling platform.

The VKontakte technology automatically creates text and distributes it in accordance with the frames, ensuring the exact appearance of subtitles at the time of speech. In addition, she knows how to place punctuation marks and put capital letters. The company promises to improve the technology for generating subtitles by adding the ability to split speech into different cues to make them easier to understand.

It works as follows. First, using intelligent noise reduction technology, the audio track is cleared of background sounds, after which the neural network recognizes the words and generates the finished text. Next, the algorithms place punctuation marks and capital letters, and at the final stage, using machine learning, the text is distributed among frames (synchronization of text with speech on video).

Currently, subtitles are available to some users in popular videos, as well as in videos from verified communities. True, so far only in experimental mode. By the end of this year, the automatic subtitling function should work in most videos on the social network.

You may also like