Benchmarking Expressive Japanese Character Text-to-Speech with VITS and Style-BERT-VITS2
This work addresses the problem of generating realistic and consistent Japanese character speech for applications like language learning and dialogue generation, representing an incremental improvement with specific benchmarking.
The paper tackled the challenge of synthesizing expressive Japanese character speech by benchmarking VITS and Style-BERT-VITS2 JP Extra on character-specific datasets, finding that SBV2JE matched human naturalness with a MOS of 4.37 vs. 4.38, achieved lower word error rates, and showed slight preference in comparative scores.
Synthesizing expressive Japanese character speech poses unique challenges due to pitch-accent sensitivity and stylistic variability. This paper benchmarks 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.