Yandex Cloud's artificial intelligence has learned to automatically place punctuation marks during speech recognition.

Yandex Cloud has updated Yandex SpeechKit, a service for speech synthesis and recognition. Users now have the ability to automatically insert punctuation marks when converting speech to text. This feature is ideal for communicating with voice assistants, automatic transcription, or generating subtitles. Text recognized by the neural network will be easy to read and as close as possible to literary style.

The Yandex SpeechKit feature is called Punctuator and can work in real time, recognizing the speech of voice assistants, as well as when recognizing previously recorded audio files.

Punctuator was developed using two sequentially operating machine learning models. The first converts speech to text, and the second inserts punctuation marks in accordance with the norms of the Russian language.

Punctuator is available to all Yandex SpeechKit users and can be configured using standard API tools. It is noted that many customers today use Yandex SpeechKit to create voice assistants with a dialogue interface.

“In such projects, it is important to make the text recognized by the machine as close as possible to how it would be written by a real interlocutor. Punctuator solves this task simply,” said Vasily Ershov, head of development at Yandex SpeechKit.