CST created a voice based on neural networks for the hero of the film "One Hundred Years Ahead"

2024: Making the Voice of the Werther Robot

The Speech Technology Cent er Group used the latest developments in the field of neural networks and artificial intelligence to create the voice of the hero of the film "One Hundred Years Ahead" - the Werther robot. The company announced this on April 23, 2024.

Millions grew up on the works of Cyrus Bulychev. We had a desire to introduce a new generation to the beloved heroes, while making this understandable to the modern viewer in the cinema language. Therefore, working on the film "One Hundred Years Ahead," we sought not only to create a visually rich world that exists according to the eternal laws of friendship, love, the struggle of good with evil, capable of captivating a wide variety of viewers - children, adolescents and adults, but also bring current modern technologies. The Werther robot was voiced by Fyodor Bondarchuk, and in order for the audience to hear not only the familiar voice of Fyodor, but also the voice in the form of a hero - a robot, some phrases were created using speech synthesis technology based on the most modern AI solutions, said Alexander Andryushchenko, director, scriptwriter.

The CST group creates world-class speech technologies, speech synthesis is one of them. In this project, we created a neural network personalized synthesis of Fyodor Bondarchuk's speech, applied technologies for cloning and modifying voice, implemented hybrid synthesis - changing and removing words, adding pauses of hesitation - hitches in oral speech. Using Voice Cloning and Voice Conversion technologies, the CST team managed to develop a speech synthesis model that allowed changing voice characteristics, influencing intonations, breaking the rhythmic picture of speech, creating unnatural, but typical effects for a robot. At the same time, we managed to realize the main thing - to preserve the original and recognizable timbre of Fyodor Sergeevich's voice, his unique style, emotions. To do this, all phrases were synthesized anew, relying on the original remarks and transferring all their characteristics, noted Dmitry Dyrmovsky, CEO of the CST group of companies.

My hero, like a real robot, was created with the participation of artificial intelligence. Thanks to colleagues from Flysound and the CST group of companies, a unique sound of this character was invented: Werther's neural network model of voice can accelerate or slow down his speech on demand, change the order of letters in words, add clicks, stuttering effects, repetitions... In general, satisfy any fantasy - while observing the rules of voice confidentiality. It was incredibly fascinating to watch it and be involved in it, told Fyodor Bondarchuk, producer, voice actor of the Werther robot.

In traditional projects for developers, as a rule, the task is to create a synthesis model that corresponds to the voice of the original as much as possible, and in this case the task was not standard, the opposite is to create a speech synthesis with a specific effect, when a voice very similar to a person forms the viewer's feeling that this voice is not real. It was also complicated by the fact that Fedor's voice was cloned from four minutes of unique material recorded in the style of the hero of the film, and the task was to introduce specific speech effects into the voice according to given characteristics - such that standard sound engineers do not work.

This picture is a joint project of the companies "Hydrogen,," Art Pictures StudioNational Media Group"," "Central Partnership with the participation TV channel "Russia" and support of the Film Fund Okko , and the studio". " Soyuzmultfilm General Partner -. Sber