The process of “teaching” AI Armenian has marked a significant milestone: for the first time in history, a robot read an Armenian text with pronunciation as close to correct as possible.

The first Text to Speech prototype in Armenian, recited in a human-resembling voice, was created on open source platform MaryTTS (Modular Architecture for Research in sYynthesis). It allowed Armenian NeuroNetwork to conduct the digital adaptation and correction of the language independently.

ICAN Development Company Director General Artak Hovsepyan has told Mediamax that the main body of work for Armenian NeuroNetwork project was launched in June 2019. An expert team of three was involved in the production. Hovsepyan himself specializes in Oriental Studies and Armenian philology.

“The task was not so much technical as philological. It required involvement of linguists, phoneticians, readers, and recording specialists. We faced certain obstacles as we couldn’t find the specialists we needed, so we decided to just start doing the project on our own. I organized the work in two stages: first we used classic methods and then we moved to deep learning. Paradoxically, we solved an issue of such complexity without relevant specialists,” Hovsepyan told Mediamax.

Photo: Beth Rosengard

According to him, the team adhered to all legal standards while working on the project. In this stage of the project they used the audio book “The Brothers Karamazov”, published on Grqaser NGO’s website and narrated by Karine Avagyan. The robot reads the first text in Armenia in her voice.

“We did it with the minimal data set. It’s a long process and we still need to enhance our capacities. First, we collect voices via audio books. We want to build a library of different voices. We also need to process them, create emotional effects so that the robot can read out any text. The computer has to be able to listen, understand and recite. The ultimate goal is to get a natural-sounding voice,” said Artak Hovsepyan.

They plan to collect up to 15 voices within a year or two and create the library within six months.

The project can be applied widely in various sectors: education, security, business, etc. According to Artak Hovsepyan, Armenia’s IT sector has developed so far by providing services to other countries and there is a lack of Armenian technologies.

“Unfortunately, neither the government nor large business entities get involved in our initiative. We only rely on ourselves and work to create opportunities and attract partners,” said Hovsepyan.

Photo: Beth Rosengard

He added that the technologies the team worked on could be compared to a sharp knife – capable of both helping and hurting.

“If someone like me can do it, many people will try to use similar technologies for different purposes. Significant issues arise in the area, so we will keep these technologies undisclosed and limited to a small circle until the government gets involved or new public agencies come forward to manage the issue. We have a need in a specialized body, comprised of trustworthy people, which will regulate the field,” concluded Hovsepyan.

Marie Taryan