Zero-shot emotional speech synthesis based on feature decoupling and adaptive loss-threshold reweighting
Yan Zhu, Yu Wang, Yijin Zhou, Bomin Liu, Rui Zhou · 2026
Read Zero-shot emotional speech synthesis based on feature decoupling and adaptive loss-threshold reweighting by Yan Zhu, Yu Wang, Yijin Zhou, Bomin Liu, Rui Zhou (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0358777.txt.
Description
Yan Zhu and four colleagues tackle a weakness in synthetic voices that must sound emotional, in PLOS ONE in 2026. Zero shot systems can copy a new speaker. Yet feelings often blur into the words. Their fix separates the two streams. Text meaning and emotional tone travel through parallel attention paths and merge only at a controlled gate, so neither corrupts the other during generation.
They also built a fine grained emotion space of sixteen classes, sorted from pitch and energy patterns in speech. During training, the system dwells on its hardest examples. Samples with unusually high loss get extra weight. Tested on speakers the system had never heard, using a standard emotional speech collection, it produced words with a low error rate and clean sound measures. The design targets expressive voices for media and assistance tools.
Catalog record
| CREATOR | Yan Zhu, Yu Wang, Yijin Zhou, Bomin Liu, Rui Zhou |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47240 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |