saya
Comparative Evaluation of Expressive Japanese Character Text-to-Speech with VITS and Style-BERT-VITS2
Rackauckas, Zackary, Hirschberg, Julia
Synthesizing expressive Japanese character speech poses unique challenges due to pitch-accent sensitivity and stylistic variability. This paper empirically evaluates two open-source text-to-speech models--VITS and Style-BERT-VITS2 JP Extra (SBV2JE)--on in-domain, character-driven Japanese speech. Using three character-specific datasets, we evaluate models across naturalness (mean opinion and comparative mean opinion score), intelligibility (word error rate), and speaker consistency. SBV2JE matches human ground truth in naturalness (MOS 4.37 vs. 4.38), achieves lower WER, and shows slight preference in CMOS. Enhanced by pitch-accent controls and a WavLM-based discriminator, SBV2JE proves effective for applications like language learning and character dialogue generation, despite higher computational demands.
Humanoids With Attitude (washingtonpost.com)
Ms. Saya, a perky receptionist in a smart canary-yellow suit, beamed a smile from behind the "May I Help You?" sign on her desk, offering greetings and answering questions posed by visitors at a local university. But when she failed to welcome a workman who had just walked by, a professor stormed up to Saya and dished out a harsh reprimand. "You're so stupid!" said the professor, Hiroshi Kobayashi, towering over her desk. "Eh?" she responded, her face wrinkling into a scowl. "I tell you, I am not stupid!"
I lost my job to a robot
Saya had been teaching for seven years. Her impressive but short CV included stints in a few rural areas, overseas and as a substitute teacher. The difference is Saya is a remote controlled robot who taught her first class of 10-year olds in 2009. While we've all heard and read the stories of manual or labour-type jobs easily replaced by robots, increasingly the jobs we previously thought safe are no longer -- teachers, bankers, data analysts and the like are all at risk. But what do we really have to fear?
Digital daughter
But "Saya" was a different kind of star, because she is the product of a Tokyo computer lab. And like all "parents", her creators have big ambitions for her, writes the BBC's Yvette Tan. "'I think I've seen her somewhere' or'She looks like someone I know' are what people usually say when they see Saya," says Yuka Ishikawa, one half of the husband and wife graphic artist team behind Saya. When the couple first posted pictures of the hyper-realistic schoolgirl online last year, it was a revelation about what can be achieved with computer design. Her slightly askew school tie, heavily fringed hair, freckled skin and teenage pout left thousands trying to work out whether or not she was a real person.