What we do with your voice

We process your voice recordings only to score how close they sound to the target and to rank coaching advice. We don't identify you from them, don't build a biometric profile, and don't train any voice-cloning model. Every take you hear back is your own voice.

Voice data only becomes biometric when it's used to identify someone. We don't do that, so your recordings are treated as ordinary personal data — but we still keep them locked down.

What we collect

• Voice recordings — short clips of you reading a dialogue (≤ 30s each).
• Pair ratings — your 1–5 scores on the rate-pairs page.
• Advice ratings — whether each coaching note helped, and any flags you raised.
• Extracted features — pitch, breath, tempo, etc., measured from each recording.
• Optional details — sex / age range / accent, if you tell us at sign-up.
• Account activity — sign-in time, IP address, browser string.

How long we keep it

Your recordings, scores, and usage data are stored for 90 days, then the raw audio is permanently deleted. We use this data to improve TrickTalk's coaching and product — figuring out which voices people try, where they get stuck, and what's actually helping. Aggregated insights (with no link back to you) may be retained longer so we can track our coaching model's improvement over time.

You can export or delete everything at any time before the 90 days are up. Deletion is immediate and removes both the audio and the linked feature rows.

Analytics we use

• Google Analytics on tricktalk.io (home + signup pages only) — traffic sources and conversions.
• PostHog inside the app — event tracking + session replay so we can see where flows break. Passwords and the username you type are masked; voice recordings are not played back in replays.

Both run only after you accept analytics cookies. You can change your mind any time via the "Manage cookies" link in the home page footer.

Independent of those, we always collect first-party measurement (which character you opened, how long you spent on a dialogue, etc.) directly to our own database. That data stays on our servers, follows the 90-day retention above, and is what powers the per-product improvements you'll see in the beta. It's covered by the legitimate- interest basis disclosed here.

Your rights

Export gives you a zip with everything — recordings, transcripts, ratings, extracted features — plus a JSON manifest describing the contents. Usually ready in seconds.

Delete removes your account and unlinks your audio. Other people's ratings that referenced your takes get tombstoned. We log that the deletion happened (timestamp + who did it) for compliance.