Couple’s Secret to a Lasting Marriage: Accepting Each Other’s Differences

Latvian television personality Sanda Dejus has described her marriage to Raivis Dejus, a volunteer leading efforts to improve speech recognition technology for the Latvian language, as a successful pairing of opposites. In a recent interview, Dejus described herself as boisterous and constantly on the move, while her husband is more reserved and focused on his work.

“It’s amazing how he can put up with me,” Dejus reportedly said, according to a report in the Latvian magazine Ieva. “I’m flamboyant, loud. I haven’t had many relationships, but guys always told me, ‘You could be a little quieter, we’re people!’ How can I be a little quieter?”

Raivis Dejus is the driving force behind “Balsu talka” (Voice Talk), a project dedicated to collecting speech data to enhance voice recognition software for Latvian. The initiative aims to enable native-language interaction with voice assistants like Alexa and Siri, currently limited in Latvian language capabilities. According to Dejus, he is not the sole contributor to the project, despite being a highly active participant.

Sanda Dejus acknowledged the contrast in their lifestyles, noting that she is often away from home while her husband dedicates his time to his IT-focused work. “In the evening we meet under the covers and cuddle sweetly. And it’s good that it’s like that, because if we were both ‘go-getters,’ who would look after our child?” she said.

The couple, married for 12 years, are parents to a daughter, Dārta. Raivis Dejus was previously recognized with a Boņuka award for his work promoting the Latgalian language, a regional dialect of Latvian. “Balsu talka” utilizes the Mozilla Common Voice platform to gather speech samples, making the data openly available for research and development of speech technologies. The project’s efforts contribute not only to technological advancements but also to the preservation of the contemporary Latvian language for future generations and linguistic study, according to reports.

Dejus noted that the collected speech data has already contributed to advancements in voice command technology, potentially enabling users to control devices like lights with Latvian voice commands. The initiative also supports the development of Latvian-language subtitles for video calls and voice-to-text applications.

As of March 22, 2026, the Mozilla Common Voice Spontaneous Speech dataset for Latvian contains 56 clips representing 0.54 hours of recorded speech, with 0.04 hours validated, collected from 8 speakers.