Hlavní stránka Info portálu ZČU

Academic painter, sculptor, restorer, and educator Jaroslav Šindelář organized an exhibition of sgraffito and paintings on the occasion of his 75th birthday, which took place on June 14 at the riding hall in Světce near Tachov. He also launched two of his books at the opening. This planned launch led the artist's wife to Jindřich Matoušek from the Department of Cybernetics and NTIS research centre at the Faculty of Applied Sciences. Renata Šindelářová wanted to surprise her husband and wanted excerpts from the books to be read by the author himself. "Mrs. Šindelářová approached me with a question whether we could use technology to restore her husband's voice to its original form, as he had undergone a total laryngectomy and communicated quietly and with difficulty," Jindřich Matoušek described the beginning of the collaboration.

It was impossible to directly convert Jaroslav Šindelář's current "silent speech" into his original voice. Nevertheless, cyberneticists created a voice model to generate speech from any text. "He will be able to write the text on his mobile phone, tablet, or computer, and the technology will then convert it into a synthetic voice that will approximate his original voice before the surgery," added Jindřich Matoušek. The experts had nine recordings from 2003 to 2021, totaling about 80 minutes. Unlike the previous experience creating a synthetic voice for Karel Gott, where the team worked with more than 20 hours of studio speech recordings, this material was recorded at home – often with noise, echoes, or distracting sounds (such as birds singing or dogs barking). These recordings first had to be acoustically cleaned using noise reduction techniques, which, however, can easily affect the tone of the voice, which was undesirable.

Due to the limited amount and data quality, the team decided to proceed differently than in the case of Karel Gott. The experts used a pre-trained model trained on a large amount of speech data from various speakers and "fine-tuned" it using carefully selected 5 minutes of Mr. Šindelář's speech. "For speech synthesis using this model, in addition to the specific text, what is to be said, we also use a carefully selected and acoustically cleaned reference sample of Mr. Šindelář's voice. It is a single sentence lasting 10 seconds," explained Jindřich Matoušek. "Human capacity is limited, and this technology will bring about a rebirth. A rebirth of the voice, which was originally a human entity. My husband was pleased with this gift," said Renata Šindelářová after the opening.

Jaroslav Šindelář's exhibition can be viewed until September 15 at the riding hall in Světce near Tachov.