The DJ's AI voice

This voice is an AI voice: a speech model (Kokoro) made on the computer that runs Reelhost. It is the one place Reelhost says so; the DJ's screens don't repeat it. Nobody has yet listened to how natural it sounds, so that is not promised.

• One button. In the DJ's voice settings, Set up the DJ's voice downloads everything in one go, with one progress bar and the total size shown before you click. Reelhost picks the voice model that fits your computer, checks every file quietly, and carries on where it stopped if the connection drops. Remove takes away everything it added.
• What it says. The artist and title of each song, the odd comment, a true fact about the song when you allow it, a call to the dance floor in a party, and guests' requests and shout-outs by name. Facts come from your own library, and, only if you switch it on, from MusicBrainz (the artist's name is the only thing sent; the answer is kept on this computer for 90 days). If there is no fact, the DJ says none.
• You decide how much it talks. How often, whether it says each song, only new ones or every few, facts, requests, call-outs, quiet hours, and the most seconds it may speak in ten minutes. A party starts with much less than a dinner. Anyone at a party can tap Quiet please on their phone to silence it at once, and guests can ask "Who is this?", "Tell me a fact", or for more or less energy.
• A few voices that fit the mix. The DJ has a handful of voices with plain names and one line each (for American English: Sunny, warm and upbeat; Ace, easygoing host; Nova, club energy; Slate, calm and late-night). Left on Fit the mix, a party gets the lively voice and a chill mix the calm one; pick one to use it everywhere. Each has a Hear a sample button. The voices were chosen without listening, so try them. Excitement follows the mix too: calmer and slower in a warm-up or wind-down, livelier at the peak, but never more talking than you allowed.
• No waiting. The lines are made ahead of time, while the song before is still playing. If one isn't ready in time, the DJ uses its recorded voice or shows the words as a caption: the music never waits for the voice.
• Languages. English comes in two kinds, English (US) and English (UK), chosen by your language setting; the UK one uses British voices and a British way of saying words, which is a rule-based approximation that no British listener has checked yet. Spanish and Italian need nothing extra: Reelhost has its own spelling rules for them. Spanish is spoken as in Spain, and Italian can't tell open from closed e and o or a third-from-last stress without a written accent, so some words will sound off. French, Hindi and Brazilian Portuguese need espeak-ng, a small separate program under the GPL licence that is not part of Reelhost; Reelhost asks you once, in one line, before it downloads it (Windows). On a Mac, Linux or Docker, install it yourself and the DJ finds it. Until then the DJ speaks English and says so. Japanese and Mandarin aren't available yet. Nobody who speaks these languages has read the DJ's lines or listened to the voice in them.
• Names can come out wrong. Words that aren't in its dictionary, many artist names among them, are said by spelling rules and can sound off. Type the name as it's written and how it should sound in How names sound, then use Hear it. The list stays on this computer.
• What the DJ learned. Fires, "more like this", "less like this" and quick skips are remembered on this computer as counts only, and fade over about six months. Forget what the DJ learned clears them.

When you remove Reelhost, the AI voice, espeak-ng if Reelhost added it, the lines the voice made and the facts it kept go with the DJ's other downloads; your choices for it stay if you keep your settings.