Yandex Cloud: Neural network for speech recognition

The main articles are:

• Neural networks (neural networks)
• Speech Recognition (Technology, Market)
• Speech technology: On the path from recognition to understanding

2022: Neural Network Launch

At the end of June 2022, Yandex.Cloud launched a neural network that recognizes 10 foreign languages ​ ​, with which you can create voice assistants and robots for call centers with large client flows.

In particular, the neural network created on the Yandex Cloud cloud platform knows English, French, Danish, Finnish, Turkish and other languages. She can recognize speech on any topic - short and long phrases, names, addresses, dates and numbers.

Yandex launched a neural network recognizing 10 foreign languages ​ ​ for robotic call centers

The system is able to recognize the language automatically in one stream: the neural network can switch between languages ​ ​ at any time. At the same time, it is possible to give her "tips" to improve the quality of recognition.

The multilingual model is based on the Transformer architecture. Transformers process input words in parallel and independently of each other, thanks to which they learn quickly. In order for the text to be coherent and grammatically correct when switching to another language, the neural network was trained on dozens of terabytes of data from professional datacets and data arrays from Yandex services and applications.

The neural network works in the speech synthesis and recognition service Yandex SpeechKit. By the end of June 2022, more than 1000 customers use the service, including X5 Retail Group and Dodo Pizza. The new multilingual recognition is available for Yandex SpeechKit user projects and is configured with standard tools in the API.

We are committed to ensuring that dialog scenarios for developing voice assistants like Alice, automating customer service, conducting surveys and quality control are available to every business. Introducing any feature into speech technologies requires large amounts of data, highly qualified specialists and hundreds of hours of training, "said Vasili Yershov, head of development at Yandex SpeechKit.[1] languages

Notes

Шаблон:Remarks