Choose a variety of voices for this video. Return JSON {"voice":"...","emotion":"...","tempo":"...","accent":"..."}: voice — voice timbre, gender and age; emotion — emotion of presentation; tempo — rate of speech; accent - language and character of pronunciation. Justify each field for the task, not abstractly.
Mechanics: speech synthesis occurs in two stages. First, the model predicts prosody - pitch, duration of sounds, pauses, intonation - then the vocoder turns this into a sound wave. The same text sounds different because the model guesses the emotion and rhythm from punctuation, letter case and specified parameters. Your levers: timbre (male/female/neutral), age, emotion, tempo, language and accent. Insider: it’s more reliable to define an emotion not with the word “cheerful”, but with a context-remark and correct punctuation - a period calms, an ellipsis pulls, an exclamation raises energy. Second insight: for advertising, use a voice 10-15% faster than comfortable - during editing, slow speech sounds sluggish. Newbie mistake: looking for one “universal pleasant voice” for everything. A voice is a casting: a narrator for an audiobook will kill a dynamic video, and a cheerful salesman will not read a meditation.