About FRACTAL

We use the service “ElevenLabs,” which enables users to clone and utilize their own voice. For identity verification, we have implemented a technology called “Voice CAPTCHA.” “Voice CAPTCHA” works as follows:

■Authentication Process
For Professional Voice Cloning, users upload high-quality voice samples (1–2 hours). They are then asked to read randomly generated specific sentences aloud.
■Verification
The system analyzes the live voice input and compares it with the uploaded samples to confirm that the voice belongs to the user and is neither synthesized nor pre-recorded.
■Safety Measures
If users attempt to bypass verification or upload a voice that does not match their own, the system is designed to treat it as a failure.
■Support
If automatic verification fails, users can retry after 24 hours or contact support to request a manual review.

For more details on FRACTAL’s collaboration with ElevenLabs in the “Soyogi Fractal” voice AI project, as well as third-party supplementary information regarding ElevenLabs’ safety measures, please refer to the links below.

• Eleven Labs Safety
• How to Protect and Use Voice Rights
• Third-Party Supplementary Materials (Daiwa Institute of Research)

Since its founding, Techno-Speech has consistently used only data obtained with the explicit consent of the original performers, as well as open-license datasets, for training its speech synthesis systems.

This policy is strictly applied across all projects, both B2B and B2C, and we place great importance on speech synthesis built on clean, properly licensed data.

The “Soyogi Fractal” project is also designed based on this philosophy. Our company was responsible for developing the voice libraries for the “CeVIO AI” and “VoiSona” series.

The Japanese talk voice and singing voice libraries are constructed solely from recordings of Mr. Yuki Kaji’s own voice, without the use of foundation models or external datasets.

For the English talk voice library, training was conducted using voice data from a total of six speakers (Japanese and English), combined with Mr. Kaji’s Japanese voice. All data used in this process is fully cleared in terms of licensing.

This approach is consistently applied across all of our initiatives. In our voice creation software “VoiSona,” we place strong emphasis on the same principles.

With the theme of “coexistence and mutual prosperity between humans and AI,” VoiSona incorporates mechanisms such as subscription-based models to ensure that voice providers—namely performers—receive continuous and fair compensation.

Moving forward, we will continue to prioritize both technological advancement and ethical responsibility, contributing to the sustainable development of voice AI.

A support message titled “For the Future of Entertainment and AI” has been published on note.

An excerpt is provided below.

"If we focus purely on the technical aspect, it is now possible to generate a voice that resembles someone’s tone from as little as five seconds of audio.
However, while this may replicate the sound of a voice, it cannot capture the subtle breathing, emotion, and intent that voice actors bring to their characters.

The AI we create is built through long recording sessions to capture the emotions and intent behind each voice.
The scripts used for recording are tailored individually for each character. In some cases, voice actors provide feedback during studio sessions to ensure that the script aligns with their interpretation of the character.

All participants approach the recording process with genuine care and love for their characters. Mr. Kaji has even offered additional recording sessions in pursuit of creating something even better. We have received tremendous support in building high-quality AI."

• note