Vovó Help

The persona

A character of the channel's, with a name, a personality and a voice, who answers chat through an AI model with your own key.

Setting up

Integrations, AI, keeps the providers' keys on this computer, the same for every profile, and says what each turns on; Settings, Persona, sets the provider, the name, the personality, the model and the voice. Without the chosen provider's key, it is asked for right under the provider, and the rest of the persona stays off until it is there.

Voices

Windows' voices are free, and more languages come from Windows Settings › Time & language › Speech. Piper's are free too, and sound more natural; they are made on this computer (below). OpenAI's voices need its key, in Integrations › AI. An action may name its own voice with vovo.say(text, { voice }).

Piper

Piper makes voices on this computer, with nothing sent anywhere. Vovó does not carry it: "Open folder", under the voice, opens Piper's folder (%LOCALAPPDATA%\Vovo\piper), and two things go in it, anywhere inside:

Back in Vovó, the voices are in every list of voices, as "Piper · …". An action names one with vovo.say(text, { voice: "piper:pt_BR-faber-medium" }). Piper reads each line in a fraction of a second.

When it answers

When someone says its name in chat, or redeems the channel points reward chosen for it. One answer per person a minute, and one at all every few seconds, so a busy chat does not spend your money.

Approval

By default, the answer waits for you to approve it: in the bell, or in AI, in the side bar. Approved, it goes to chat, is spoken on the live and reaches actions as persona.answer.

Its own account

In chat, the persona's answers are written as you, unless it has a Twitch account of its own: in Integrations, under Twitch, "The persona's account" signs that account in with a code, as yours was. Open the link in a private window where Twitch is signed in as the persona's account, or the code would connect your own again. From then on its lines are its own, and it never answers them. Make that account a moderator of your channel, so slow mode and followers-only chat do not hold it. It only writes: it reads nothing and manages nothing. An action that gives the persona a voice can write through it too, for what the persona says that is not an answer, such as a line typed for it; while the account is not connected, such a line is not written.

Memory

The persona carries a conversation on: the last few exchanges go with each question. That is all it knows, and only while Vovó is open, unless you turn on "Remember between lives" in Settings > Persona. Then the conversation is kept on this computer and, as it grows, its oldest exchanges are read once more by the persona's model, which takes the lasting facts out of them: who showed up, a nickname, a joke that caught on. That reading is one more call to the model, about one in three answers, paid with your key. What comes back grows stronger and stays; what was said once fades in a little over a month. Pin a fact and it stays for good; type one yourself and it comes in pinned; in the AI tab, the pin on an answer keeps what the persona said. Everything it remembers is listed in Settings > Persona, to forget one by one or all at once, and none of it leaves this computer except to its model, with each question. An answer you discard is not remembered. It holds 40 facts at most.

Vovó's rewards

In Integrations, with Twitch connected, make channel point rewards from Vovó itself: title, cost and whether they ask for text. It can pause, resume and delete the ones it made. Their redemptions wait in the reward's queue: when the persona answers or the text is read, the redemption is approved; when that is discarded, the points go back to whoever redeemed. Twitch lets only the app that made a reward change it, and a channel holds 50 at most.

Read a reward aloud

In Settings, "Voice", choose a channel points reward by its title: what the viewer types in it is read on the live, in a voice of the profile's. It always waits in the bell for approval first (here, or from a device allowed to approve), and web addresses are left out of the reading. A chat message is read the same way from the speaker beside it, in the side bar's Chat or on a phone allowed to read the chat: your click is its approval, so nothing is read by itself. A line the persona's own account wrote is the persona's, and an action that gives the persona a voice says it; actions hear every reading as tts.spoken.

Speech to text

What you say into the microphone becomes text: for captions an action shows (speech.text, a phrase at a time, a second or two after it was said) and, with "Hear me" on in the persona's settings, for the persona to answer when you say its name. The microphone opens only while something needs it, and the status bar shows it as it shows the other readings; nothing is kept.

Two engines write it down, chosen in Settings, "Voice":

"The live's words" are names your live uses (the channel, the persona, games, friends), separated by commas: they come back spelled right. Keep it a list, not a sentence: Whisper reads them as words just said, and a sentence invites it to carry on with one. A phrase that is only those words coming back is thrown away, as are the lines engines make up when they hear no words.

The persona hears its name at a pace: 15 seconds between answers at least, and 30 an hour at most. Its answers wait for approval as any other, and its own voice heard back through the microphone is no question to it. Talk with headphones, so the microphone hears you and not the stream.