«VKontakte» launched the deciphering of video messages in the VK messenger. Three neural networks are involved in the work of the function

The deciphering technology is fully automated and operates on the basis of machine learning methods developed by the VKontakte team. For high-precision decryption, video recordings are processed by three neural networks: they recognize sounds, separating them from the video sequence, form words from the sounds and define sentence boundaries.

The technology is trained on movie subtitles and is able to understand incoherent speech and informal lexicon, remove pauses from the recording, divide the text into sentences and semantic blocks. Speech recognition is fully automated and takes place without human participation.