A speech recognition model in Russian has become available in Russia

The Artificial Intelligence Center of the T-Technologies Group has made the T-one streaming speech recognition model publicly available. With a relatively small size, it leads in the quality of recognition in Russian on noisy and compressed recordings from call centers, the developers said.

The model was originally designed to work in real time with an audio stream of any length. The technology provides low latency, which is critical for voice assistants, call center automation, and call analytics, the developers noted.

Streaming speech recognition allows you to process calls in real time, improve the speed and quality of service, as well as reduce the burden on operators and staff costs, the developers emphasized.

T-one is distributed under the Apache 2.0 license, which allows free commercial use and modification. The weights of the model and the code for running in a high-load environment and adapting to your own tasks are available on special websites for software developers.