Introducing NeuraVoice2 Beta: Ultra-Fast, High-Quality Text-to-Speech Neural System
In 2017, we initiated research on end-to-end speech synthesis using deep learning technology. Now, after numerous upgrades, we're pleased to introduce NeuraVoice2 Beta. It's set a new benchmark for GPU inference in the Real Time Factor (the ratio of the time length of speech data to the time taken for synthesis), achieving an impressive 0.0227.
For more information, please refer to these articles:
NeuraVoice2 Beta offers several outstanding features:
・High-speed, high-quality speech synthesis with a Real Time Factor (RTF) of 0.01~1.1*1.
・Accurate reading of numbers up to 1,000 trillion digits, including special notations such as superscripts.
・Adjustable speech speed, pitch, and intensity.
・Advanced settings for stemming, reading symbols, eliminating stop words, and reading keywords only.
・Compatibility with the same algorithm across multiple languages.
・Customized speech synthesis with the preparation of original training data to capture a specific speaker's voice or certain group of words.
We are excited to incorporate this product into CITYWALK, a voice guidance system currently under development by ThinkX.
Primarily, we expect to offer NeuraVoice2 Beta to customers requiring high-speed speech synthesis, particularly in edge devices with limited communication environments.
For more information or product introductions, please contact us at inquiry@thinkxinc.com.
