DataIsland Software
← All Tutorials

Talent — Tutorials

Written guides for every feature in Talent, the voice-library app in the PolySpeak pipeline. Work through them in order for a full introduction, or jump to any section using the contents list.

1 Training a Voice from a Recording

What you'll learn: turning a source recording into a trained AI voice.

  1. Click New Voice in the sidebar.
  2. Record directly in the app, or import an existing recording — roughly 10 to 30 minutes of clean source audio gives the best results.
  3. Name the voice and set the number of training epochs.
  4. Start training. Talent runs the training pipeline entirely on-device — this can take a while depending on your Mac and the recording length.
Clean audio in, better voice out. A quiet recording with consistent mic distance and minimal background noise trains a noticeably cleaner voice than a noisy or inconsistent one.

2 Importing an Existing Model

What you'll learn: adding a voice you've already trained elsewhere, without retraining it.

  1. Click New Voice, then choose Import Existing Model….
  2. Select an RVC .pth model file, and its matching .index file if you have one.
  3. Choose the RVC architecture version the model was trained with (v1 or v2).
  4. Talent registers it as a new voice in your library, ready to cast.
No training history for imported voices. Since Talent didn't train it, an imported voice's origin is recorded honestly as "imported" rather than showing fabricated training details.

3 Reviewing the Preview Sample

What you'll learn: hearing what a freshly trained voice actually sounds like before trusting it.

  1. After training finishes, Talent converts a short clip of the original recording through the new model.
  2. Play the preview — this is a real converted sample, not a placeholder, so it's an honest test of quality.

4 Consent Tracking

What you'll learn: recording consent for a trained voice, and why it matters.

  1. When training a voice from someone else's recording, complete the consent attestation step.
  2. The consent record is stored alongside the voice's other data, viewable from the voice's Authenticity section.
This is your responsibility, not just a form field. You must have real consent from the person whose voice you're recording, or use only your own voice. Talent's attestation records that you've confirmed it — it can't verify consent actually exists. See the Terms of Use for the full policy.

5 Catalog Numbers & Casting Status

What you'll learn: keeping a growing library organized and knowing which voices are production-ready.

  1. Every voice gets a permanent catalog number automatically when it's added to your library.
  2. Toggle a voice between Testing and Ready for Casting from its detail view once you're confident in its quality.
Leave new voices as Testing at first. Cast them in a scene or two, confirm the quality holds up, then flip to Ready for Casting so you and any collaborators know it's trusted for real production work.

6 A Voice's Resume

What you'll learn: how a voice's performance history builds up automatically.

  1. Open a voice's detail view and scroll to the Resume section.
  2. Every credit here was added automatically — by TableRead when it generates a full read-through, and by Studio when it exports a finished take.
No manual entry needed. A voice's resume fills itself in as you use it across the suite — there's nothing to type in yourself.

7 Multi-Window

What you'll learn: comparing voices side by side.

  1. Press ⌘N, or choose File → New Window, to open a second library window.
  2. Select a different voice in each window to compare them side by side.