The Voice Replication feature allows users to create personalized AI voice models with the following implementation mechanism and operational requirements:
Technical Principles
Based on the deep learning speech synthesis technology of KDDI, it analyzes the voice samples provided by users, extracts the voiceprint features (e.g., timbre/tone/pronunciation habits, etc.), and ultimately clones a personalized voice with a similarity of 90% or more.
material preparation
- recorded text: Required to read aloud the training text specified by the platform (usually containing 100-200 sentences)
- audio quality: Recommended to record in a quiet environment using a professional microphone with a sampling rate of ≥16kHz
- Content Coverage: The text should contain commonly used words, polyphonic words, and specific combinations of pronunciations.
application scenario
The cloned voice can be used for: audiobook reading, personalized video dubbing, brand exclusive voice logos and so on. This feature is especially suitable for knowledge bloggers, education and training workers and other user groups who need to maintain voice consistency.
It should be noted that for ethical reasons, the platform requires that voice cloning must be authorized by the person themselves, and that they may not copy another person's voiceprint.
This answer comes from the articleCyberSmart: Converting Text to Speech and Digital Human VideoThe
