A donation message read by a stock robot voice is fine. The same message read in your own voice is a moment your community talks about for weeks. This guide explains how streamers clone their own voice for TTS, what happens to your voice data, and what stops viewers from making you say things you never would.
One thing first: ControlPlay TTS is enabled per account during the current beta, and Own Voice, the AI clone of your own voice, is a separate, invitation-only option. Library voices work without it.
How TTS in your own voice works
- You record short samples. In the ControlPlay client, you upload clips of your own voice: up to 25 samples, each up to 10 MB, in common formats like MP3, WAV or OGG.
- An AI voice model is created. The samples are sent once to the voice provider to create your voice model. There is no training queue; the voice is available right after creation.
- Viewers write, your voice speaks. A viewer submits a message through the action you enabled in the Twitch Extension, using available Coins. After moderation, the message plays in your cloned voice.
You choose the voice per TTS action, so you can offer library voices and your own voice side by side, at different Coin amounts.
Is it safe to upload your voice?
The questions that matter are ownership, storage and deletion, so here is the plain answer for ControlPlay:
- Only your own voice. You may only clone yourself. The client shows privacy information before you create the clone.
- Samples are not kept. Your recordings are transmitted once to create the voice model. Afterwards, generating speech only sends the voice reference and the message text.
- You can delete it. Remove your cloned voice in the client at any time; it is deleted at the voice provider as well. Deleting your ControlPlay account removes it automatically.
Cloning someone else is a different story. A celebrity or another streamer never consented to your upload, and impersonation is covered by Twitch's Community Guidelines. Keep cloning where it belongs: your voice, your consent, your channel.
What stops viewers from abusing your voice?
The honest fear behind every custom voice: a stranger typing a slur and your own voice saying it. ControlPlay checks every message before any audio exists: a word filter catches known terms and the usual bypass tricks, then a semantic AI check reads the full message for hate, harassment and threats. Messages that fail are never synthesized, and in block mode the viewer's Coins are returned automatically.
You additionally control casual profanity, your own blocked words and per-viewer timeouts. The TTS moderation guide covers the complete system. And to be precise about the mechanics: viewers never get access to your real microphone. They submit text; your configured, moderated pipeline does the speaking.
Your voice in Discord and game voice chat
Played on the stream output, a cloned-voice message is a normal TTS moment. Routed through the configured virtual microphone, it reaches Discord, TeamSpeak or a compatible in-game voice chat, so your teammates hear "you" react to chat mid-match. Whether a specific voice chat picks it up depends on your audio setup, so test it in a private call before going live.
ControlPlay speciality: viewer-written lines, spoken in your own cloned voice, moderated before playback, and routable into voice chat. Each piece exists somewhere; the combination is what makes clips.
Recording samples that sound like you
The model can only be as good as the audio you feed it:
- Record in a quiet room without music, game audio or hum.
- Speak naturally, the way you talk on stream, not like a newsreader.
- Use several shorter clips with varied sentences instead of one monotone take.
- Skip clips with other voices in them; the model should only ever hear you.
If the first result sounds off, delete the voice and rebuild it with cleaner samples.
Set up your custom TTS voice
- Create a ControlPlay account, connect Twitch, install the Windows client and activate the Extension.
- Set up TTS with a library voice first; the Twitch TTS setup guide covers voices, routing and testing.
- Own Voice is invitation-only during the beta; contact ControlPlay support to ask about access.
- After the invitation, record your samples, review the privacy information and create your voice.
- Configure the action's Coin amount, cooldown and moderation, then test with a harmless message from a viewer account.
Start with the moderation settings stricter than you think you need. Your own voice saying the wrong thing hits harder than any robot voice ever could.