Microsoft has increased the realism of computer simulations of human speech (PHOTO)

В Microsoft повысили реалистичность компьютерного симулятора человеческой речи (ФОТО)

The Microsoft engineers managed to improve the quality of computer voices that mimic both male and female speech.

This was reported in the Microsoft Azure blog.

To overcome the difficulties of the traditional model, the researchers used a neural network for speech synthesis. Microsoft promises to provide support for 49 languages, as well as the possibility of creating unique voices for the needs of companies in the near future.

Speech synthesis using neural networks involves the comparison of the stress and length (the so-called prosodic) units of speech and their synthesis in a computer voice.

In traditional speech synthesis prosodic divided into acoustic and linguistic analysis, run various models. Causing it turns out noisy and indistinct.

But Microsoft said that they managed to combine two models of neural synthesis, which enabled to achieve greater realism in computer speech.

The developers are convinced that speech synthesis using neural networks will make it more natural communication with virtual companions and assistants. Moreover, it will provide an opportunity to convert e-books into audio books and also change sound built-in car navigators.

The computing power of Azure is available in real time, and is responsible for the Azure Kubernetes.

The simultaneous application of neural speech synthesis along with traditional talking about expanding and improving the availability of service. But so far the system has two voice options – women’s by the name of Jessa and male named Guy.

In their studies of face recognition technology and speech synthesis Microsoft competes with Google. At the end of August 2018 Google has updated its Cloud services and has announced the release of a stable API for Cloud-to-speech Text-to-Speech with the experimental function of your profiles and support for multiple new languages. At the same time service for decoding audio Cloud Speech-to-Text have learned to recognize different speakers and to determine the language of a recording of several possible.

Also, earlier, Microsoft has removed the installer from Google Chrome’s own app store. In this case, the Corporation did not specify the specific reason, saying only that the application must be “unique and distinctive”: “We have removed Google Chrome Installer from the Microsoft Store, since the application violates the policy of our store.”

Leave a Reply

Notify of