Yandex SpeechKit learned to synthesize and recognize speech in Uzbek language
The Yandex Cloud platform has developed neural networks that can recognize speech and play it back in Uzbek. Companies both in Russia and Uzbekistan can already use the new language in the Yandex SpeechKit service to create voice assistants, call center automation and speech analytics.
Customizing synthesis
Speech synthesis models are trained on a pre-recorded voice of a real speaker. At the same time, they are able to synthesize speech not only from the text, but also from its phoneme record, which specifies all the features of sounds in words. This will allow clients to correct the synthesis pronunciation of individual words - complex names, names of goods, borrowed expressions. To do this, they will need to specify the phoneme breakdown of the desired word in the text using special syntax.
Large datasets for speech recognition
Just like for speech synthesis, for recognition neural networks learn from examples. To train them, Yandex Cloud experts have collected a dataset with thousands of hours of audio and their transcripts, including short and long phrases, as well as names, addresses, dates and numbers.
Uzbek Latin
Neural networks for the synthesis and recognition of Uzbek work with the Latin alphabet, which has been used in Uzbekistan for more than 20 years. The algorithms had to learn some of the sound features of the letters. For example, the letter "X" denotes a hard [x] in the alphabet, but in foreign words can be pronounced as [xs].
"Scenarios related to artificial intelligence, particularly speech technology, are now actively developing in Uzbekistan. One of our partners estimates that the speech technology market in Uzbekistan can reach up to 395 million minutes of processed speech per year. Appearance of the new sixteenth language in Yandex SpeechKit is a big step for creating new dialog scenarios for both Uzbek and Russian companies," - said Grigory Atrepiev, head of products in Yandex Cloud.
The models are available on the website and can be customized using standard API tools. Synthesis and recognition of the Uzbek language with the help of Yandex SpeechKit is already used by a number of companies, including Kolesa Group, for example. Partner companies including TargetAI, JustAI, Unistar Digital and Tomoru are also already testing and integrating the technology into their solutions.
In today's era of digitalisation, businesses in all sectors are facing new challenges. Competition is intensifying, customers are becoming more demanding and technology is evolving at breakneck speed. To remain competitive, businesses need to change. And one of the key steps along the way may be migrating to the cloud.
The right choice of a business partner is one of the main tasks of any business. The quality of goods or services that the company receives, as well as its reputation, depend on the reliability of the supplier. To make the process transparent and convenient, scoring models for assessing reliability come to the rescue.
Summer is coming to an end, but life in the analytics field is bustling. New vendor announcements, promising projects, and the pervasive penetration of AI into all areas of our lives are paving the way for the upcoming business season.
Maksim Karankevich, Director of Data and Digital Transformation, Ultramar, in his presentation shared his experience of implementing a cellular-based industrial radio network.