- Speech dialogue
- Media transformation
- Media generation
RECAIUS™ HMM-based speech synthesis technology
Improves naturalness and speaker similarity of synthesized voice using a speech synthesis method based on statistical parameter selection.
- Achieves smooth and natural synthetic speech by introducing new acoustic feature parameters enabling precise reproduction of the speech waveform shape and through our original speech synthesis method that combines a processing to select acoustic feature parameters from among huge numbers of candidates with a statistical parameter generation method. This enables speech synthesis with dramatically improved quality and better reproduction of the texture of the voice. Furthermore, a new method that pursues even more natural voice is under development.
Applications
- Reproduce the voices of past voice actors and characters (reproduce speaker characteristics).
- Natural synthetic speech for spoken language (speeches for announcement or translation)
Benchmarks, strengths, and track record
- November 13, 2019: 2019 Kanto Regional Invention Award (Honorable Mention): High quality, low calculation cost speech synthesis technology (Patent No. N085700) (Japan Institute of Invention and Innovation) (in Japanese)
- Rapid response to a customer across manufacturing, sales, and engineering: “ToSpeak™ Gx NEO” speech synthesis middleware adopted in the major hit product “POCKETALK®” (Toshiba Digital Solutions Corporation) (in Japanese)
- “ToSpeak™ Gx Neo”, a high quality speech synthesis that can faithfully reproduce characters’ voices, supports a talking app with speech technology. (Toshiba Digital Solutions Corporation) (in Japanese)
Inquiries
Toshiba Digital Solutions Corporation
ICT Solutions Division, RECAIUS Engineering Dept. (in Japanese)
References:
- DiGiTAL T-SOUL Vol. 30:
RECAIUS™ supports workstyle innovations and richer lifestyles; A digital society that connects people and AI
A case study of “ToSpeak™ Gx NEO” middleware application: Natural and friendly synthesized voice close to original speaker’s voice (Toshiba Digital Solutions Corporation) (in Japanese) - Toshiba Review, Vol. 75, No. 5 (September 2018)
Special Feature 2: Workstyle innovations targeted by Toshiba Communication AI “RECAIUS™”
Speech middleware that can be used to build speech interfaces for everyday devices (PDF) - Toshiba Review, Vol. 71, No. 5 (September 2016)
General paper: Next-generation speech synthesis technology with enhanced voice quality and speaker similarity based on statistical parameter selection (PDF)
