Why Nixi's voice is recorded, not generated
There are two ways to make an assistant talk. You can generate speech on the fly, so it can say anything, or you can record someone saying the things you want it to say. Nixi uses recordings.
One voice, one quality
A generated voice is a different quality of voice depending on what it is saying, how long the sentence is, and how fast your PC is. A recording sounds the same every time. Nixi's voice is a single British voice, and everything it says is that voice: "Done.", a greeting that fits the time of day, "Your timer is up."
Instant, offline, private
A recording is just a sound file, so there is nothing to wait for. It plays the instant the answer is ready, with no internet and no model running. Nothing about what you asked is sent anywhere to make it speak.
Times and numbers are built from words
Nixi can tell you the time, the date, or what a timer is set for without a generated voice, because we recorded the building blocks: the numbers, the months, "o'clock", "percent". "It's three forty-five P M" is those words joined together. It takes a fraction of a millisecond.
It only says what it has a recording for
This is the part people notice. If Nixi has no recording for an answer, it does not read the answer aloud in some other, different-sounding voice. It shows the answer on the pill and stays quiet. We think a short, consistent voice that knows when to stay silent is better than a voice that can say anything and sounds like a different person each time.
Fun on purpose
Because the lines are chosen, they can have a personality. Pick "Sir" or "Ma'am" when you set up Nixi, and it uses it now and then, not every time. Ask "Are you there?" and you get a proper answer.
What comes later
Free-form talking, an assistant that can answer any question out loud, needs a voice that can say anything, and that is a different kind of thing from a set of recordings. That is where bring-your-own-key fits: you connect your own account with a voice service, and with a language model service if you want one, and choose the voice you like. It is a plan, not something you can use yet, and it will be a separate, optional path. The free app keeps its one recorded voice.