Plan and timing

Roadmap

We're in formation. The first job is to establish the project and gather consented voice recordings. Training and a public voice come after that.

Where we are

The project is in formation. We're preparing the public home, the docs, and the consent process. Recording has not opened yet.

Step 1 of 5: Formation

  1. Formation
  2. Recording
  3. Dataset
  4. Training
  5. Release

Updated 2026-11-16.

The five stages

  1. 1 · Formation — current Set up the project's home and rules. Docs, consent, licenses, the public site, and (when they're ready) a project email and the sign-up and upload forms. Recording is not open yet.
  2. 2 · Recording Collect consented Santa voices. Adults record the core set plus one line pack. Main recording window: December 26, 2026 through January 2027, after the busy holiday week. Earlier recordings are welcome once submissions open.
  3. 3 · Dataset Review and publish. Maintainers listen, run quality checks, and prepare an open dataset under CC BY 4.0. First milestone: about 1 hour of accepted audio from about 12–15 people, with every pack covered by at least two speakers.
  4. 4 · Training Build the high-quality voice. The plan is a blended community Santa, not a copy of any one person, using StyleTTS2 as the main approach. Compute and training wait until the recordings are in. Details are in the training plan.
  5. 5 · Release A free Santa voice anyone can use. A good v1 aims for about 2–4 hours of accepted audio from about 25–50 people, in late winter or spring 2027. A smaller on-device voice can come later. The model's license is still being decided.

Timeline

  1. Now · mid-November 2026

    Awareness and formation Build the public home, finish drafts, and (before any public ask) set up the project email and forms. A soft start can raise awareness, framed as something people may do after their season if they're busy now.
  2. December 15–24, 2026

    Go easy Mid-December is many Santa performers' busiest week. We won't chase people then. A quiet listing of the sign-up link is enough, once it exists.
  3. December 26, 2026 through January 2027

    Main recording window The holiday is just over, and many people have more time. Goal: about 1 hour of accepted audio from about 12–15 contributors, enough to start training experiments.
  4. Late winter into spring 2027

    Toward a good v1 Keep accepting careful, consented recordings rather than rushing. Target: about 2–4 hours from about 25–50 people. Training happens when the data is ready, not before.

Still open

  • The model license (OpenRAIL-M style is recommended, with Apache-2.0 as a fallback). See the model license notes.
  • The project email (support@opensanta.org) is set up. The project Google account, sign-up form, upload form, and private Drive archive have not been created yet.
  • A lawyer's review of the consent form and licenses before launch.

A longer, more technical plan lives in the training plan and in the repo roadmap notes.