You are currently viewing VKontakte has improved the technology for recognizing voice messages

VKontakte has improved the technology for recognizing voice messages

VKontakte has reduced the time for decrypting audio messages. The possibilities of the new recognition algorithms for Rusbase were discussed in the press service of the social network.

A neural network created by the VKontakte research team is responsible for understanding speech, including Russian slang. So that the translation of long recordings into text does not affect the speed of decryption, the developers have found a new technological solution.

Recognition of a recording up to two minutes (this is 99% of all audio sent by VKontakte users) will now take a second, the technology development team assured.

“Voice Activity Detection algorithms automatically break long audio into short blocks, decode them all at the same time, and then combine them back into one message. At the same time, they know how to divide the recording without breaking words into parts, so as not to distort the meaning of what was said, “- explained in” VKontakte “.

Recognition of voice messages is fully automated. To train neural networks, audio is used, which is specially recorded for this task by the participants of the VK Testers program.

It also became possible to record voice messages up to an hour.

VKontakte launched the recognition technology in June this year. According to the social network, from that moment on, the audience that uses voice messages has increased by 10% and is 33 million people a month. On average, each user of the service reads 8 transcripts per day.