Avatar Training Privacy Notice (draft)
Last updated: [to be filled in on the date of publication] Version: 1.9 日本語
Avatar Training (the “Service”) is a service for practising interviews, such as investigative interviews, with AI-driven characters (avatars), for training and research. This notice explains what personal information the Service handles and how it is handled. This is the English version of the notice. It is also available in Japanese.
1. Who this notice covers
- Users: people who have an account on the Service and create avatars or interview settings (including people with view-only accounts)
- Participants: people who take part, through a participation URL, in training or research conducted by a user
- Face providers: people who provide photos or videos of their face so that an avatar can be created (separate information and consent apply; see 10)
2. Operator and the person responsible for each study
- Operator of the Service: Shumpei Haginoya (operating the Service as an individual; affiliation: Faculty of Psychology, Meiji Gakuin University) Address: Haginoya Laboratory, Faculty of Psychology, Meiji Gakuin University, 1-2-37 Shirokanedai, Minato-ku, Tokyo 108-8636, Japan Contact: haginoya@psy.meijigakuin.ac.jp
- Information about users’ accounts: the operator is responsible for it (the “controller” under the GDPR).
- Information about participants: the user who conducts the study (and their institution) is responsible for it (the “controller” under the GDPR). The operator holds and processes the information on the user’s instructions (a “processor”). Please direct questions about participants’ information first to the contact given in the consent form for that study.
3. Information we handle
Users
- Email address, name, affiliation (if entered), role (user, view-only, etc.)
- The date and time you confirmed this notice, and its version
- The time zone for times on screen (if you choose one)
- Login records (times of login) and records of actions (who did what and when, such as sharing, changing settings or deleting records)
- Usage (such as the number of AI classifications and the number of characters of speech synthesis, for managing costs)
- Avatars, study settings and project API keys created by users (API keys are stored encrypted)
- Records of the test interviews you run on the test page (questions and replies, and the time zone of the device when the interview starts). They are kept linked to your account so that you can save your own interviews as CSV
Participants
- What participants enter on consent screens and questionnaires (such as their name or background information; what is asked differs from study to study)
- Interview records: the participant’s questions (as text), the avatar’s responses, and the course of the interview
- Records of how the participant moved through the pages: the pages shown and their order, pages skipped because of conditions, scores calculated from the answers, and the time at which the AI notice was shown
- Results of automatic coding (such as the classification of questions)
- Voice: when a participant asks questions by voice, the voice is sent to a speech recognition service (see 6) to be converted into text on the spot. The Service does not store voices (it stores only the questions converted into text).
- Technical information: the address the participant connects from (counted only temporarily in the server’s memory, to stop large numbers of requests in a short time; it is not kept in any record). When a participant asks questions by voice (including getting the microphone ready for it): the state of the recording (such as its length, the sound level, and the state of the microphone and the page), and the general type of the device, browser and operating system, and the versions of the browser and operating system (for example, “iPhone, iOS 18.6, Safari 26.0”; the device model is not included). These are kept in the interview record, to find the cause when recording does not work. The time zone of the device (for example, Europe/Vilnius) and its offset from UTC, when the participation starts and when each interview starts. These are kept so that the participant’s local times can be shown in the records and in the exported data.
Face providers
- Photos of the face, or short videos recorded of it (facial expressions and head movements)
- Records of consent (date and time, whose face it is, name [optional], the version of the information given, retention period)
4. Purposes of use
- Running the interview practice pages (classifying questions, choosing the avatar’s responses, synthesising and recognising speech)
- Enabling users to conduct research and analyse the results (exporting records, automatic coding)
- Creating avatar videos (from images and videos of faces; see 10)
- Operating the Service safely (login, preventing misuse, dealing with failures, making backups)
We do not use information for advertising and we do not sell it. The providers of the AI services do not use this information to train AI (see 6).
5. Where information is stored
For each project, the user chooses whether information is stored in the EU or in Japan. Participants’ information and images and videos of faces are stored and processed on the server in the chosen location. Account information (such as email addresses) is managed on the server in the EU.
6. External services (processors)
The Service uses the services of the following companies. We have concluded agreements on the handling of personal data (data processing agreements, DPAs) with all of them.
| Service | Purpose | Information sent | Storage |
|---|---|---|---|
| OpenAI (United States) | Classifying questions, automatic coding | The text of questions and responses, and the classification instructions | Not used to train AI. Kept for up to 30 days to monitor for misuse. Processed in the United States, including for EU projects |
| Microsoft Azure (speech) | Speech recognition, speech synthesis | Voice (recognition), the text of responses (synthesis) | Not stored in real-time processing. Processed in Japan (Japan East), including for EU projects (see 7) |
| Resend (United States) | Sending emails: login codes, invitations, confirmation of a change of email address and a notice of the change (to the previous address), sharing notices (also to addresses that are not registered yet), confirmation and notices of a change of system administrator, notices of a change of the system administrator’s email address (when it is changed in the server settings; to the previous and the new address), test emails to check the sender address (to an address entered by the system administrator), and operational notices (usage limits, heavy load on interviews, failed video creation, low disk space, software update requests, errors and similar; to the system administrator or to the notice address set in the server settings) | The email addresses of users and of the people items are shared with, users’ names, the addresses that test emails and operational notices are sent to, codes, and the text of the notice (which may include the names of shared items and projects, email addresses before and after a change, the email address of the person who requested a software update, and participation IDs) | Records of sending |
| Hetzner Online (Germany) | Server for the EU site, and storage of encrypted backups | All data of EU projects | Stored within the EU (Germany). Voice processing and the classification of questions are as described in the rows above |
| SAKURA internet (Japan) | Server for the Japan site (once the Japan site has started) | All data of Japan projects | Stored within Japan |
Images and videos of faces are not sent to external AI services. Avatar videos are created on the Service’s servers or on the user’s own PC (when the Studio is used; see 10).
7. Transfers to other countries
When information is sent from the EU to outside the EU (for example, the classification of questions at OpenAI or the sending of emails by Resend), this is done on the basis of the EU Standard Contractual Clauses or the EU–U.S. Data Privacy Framework. Voices in EU projects are processed in Japan (Microsoft Azure Japan East). Japan is a country that the EU has recognised as providing an adequate level of protection (adequacy decision). For participants in Japan, the text of questions and responses is processed by OpenAI in the United States. The United States is not a country that Japan’s Personal Information Protection Commission has recognised as having a level of protection equivalent to that of Japan. With OpenAI, we have agreed in a data processing agreement that the information will not be used for training, and on security measures. Japan and the EU recognise each other as providing an adequate level of protection.
8. Retention periods
| Information | Retention period |
|---|---|
| Participants’ records (entries made at consent, interview records, results of automatic coding) | The period the user has set for each project (1 month to 10 years, or indefinitely). Deleted automatically when the period has passed |
| Original images and videos of faces | The period the user has set, counted from when the video has been created (1 month by default; from immediate deletion to indefinitely) |
| Avatar videos and motion templates that have been created | Until the user deletes them |
| Users’ account information | Until the account is deleted |
| Backups | 14 days on the server and 90 days in a separate location (Germany for the EU site). Backups are stored encrypted |
9. Security measures
- All communication is encrypted (HTTPS)
- Users log in with a one-time code sent by email
- The participant endpoints can be used only by people who started from a participation URL
- API keys and backups are stored encrypted
- Information is kept separate for each account, and only the people it has been shared with can see it
- Actions such as sharing, changing settings and deleting records are recorded
10. Images and videos of faces
When the Studio is used (processing on the user’s PC)
Users can use “Avatar Training Studio” (the “Studio”; previously called the “helper app”), an app that creates the avatar’s motion templates and videos on the user’s own PC. When the Studio is used:
- The original images and videos of faces are handled only on the user’s PC and are not sent to the Service’s servers or to external services. Only the resulting motion templates and avatar videos are sent to the server
- The Studio stores the original images and videos of faces on the user’s PC and, by default, deletes them automatically after one month (the user can change the period). The user can select and delete them at any time from “保存したデータ” (Saved data) in the Studio
- The Studio accepts jobs only from pages of the Service opened on the same PC that the user has allowed
- The components and models that the Studio uses for its computations are downloaded to the user’s PC from their official distribution sources. No face data is sent when this happens
When you are asked to provide images or videos of your face so that an avatar can be created, the user who asked you to be filmed or to provide images gives you separate information at that time (the purpose, how the images are processed, the retention period, who will use them, that taking part is voluntary, and how to withdraw) and confirms your consent. The consent is recorded in the Service. Child avatars use the faces of people who do not exist (faces created by image generation, for example).
To withdraw your consent or to make an enquiry, please contact the user who asked you to be filmed or to provide images. The user deletes the original images and videos, and the avatar videos and motion templates made from them, using the functions of the Service.
11. Notice about AI
The avatars of this Service are AI-driven characters. Their responses are chosen by AI from responses prepared in advance, or are generated by AI. The video of an avatar made from a real person’s face is video that has been altered by AI. On the participants’ interview page, when the interview page is opened, the message “The person you are about to interview is an AI-powered avatar.” is shown over the avatar’s video. If the avatar is asked during the interview whether it is an AI, it acknowledges that it is an AI-powered avatar and then returns to its role. In addition, the user who conducts the research or training tells participants about this in the participation information (such as the information sheet or the consent page) before the interview begins.
12. Your rights
- You can ask to see, correct or delete your information
- You can withdraw your consent at any time (participants’ records, images and videos of faces, and the avatar videos created from them are then deleted)
- You can ask for the processing to be restricted, or object to it
- In the EU, you can lodge a complaint with the data protection supervisory authority of your country. In Japan, you can consult the Personal Information Protection Commission.
Users can correct their own name, affiliation and email address on the “Users” page, and can delete their own account (except for the system administrator’s account). Deleting an account also deletes the user’s projects and the avatars, study settings and participant records in them, as well as the images and videos of faces and the motion templates that the user imported.
Participants should contact the contact given in the consent form for the study; face providers should contact the user who asked them to be filmed or to provide images; and users should contact the operator (see 2).
13. Cookies
We use only the cookies needed to keep users logged in and, on the regional sites (EU and Japan), a cookie that keeps the chosen time zone the same on both sites (kept for one year and removed when you log out) (users only). We do not use cookies for analytics or advertising. Fonts and other components are not loaded from external sites.
14. Changes to this notice
When we change this notice, we announce it on this page, and we also inform logged-in users of major changes by email. When the version changes, you will be asked to confirm the changed notice the next time you log in. After logging in, you can view this notice at any time from “Help” at the top of the page. This notice also has a Japanese version; pages in languages other than Japanese show this English version.