Whose voice reads the line is up to you — pick it in DOTV's voice dialog: a few hundred system voices across three tabs, a row of filters, and the ones you use most favorited so they're right there next time. The way in is the pill with the swap icon on the audio input bar, which shows the voice you're on right now. This section covers how the three tabs and the filter row work, what's on a row, and what happens after you pick.
The dialog splits into three tabs:
Top right has Search (by name) and Filter (by language and gender). The list is paginated, with the pager at the bottom.
Pick an accented voice (Cantonese, Japanese, Korean and so on) and Language in the settings panel moves with it — swap the voice without the language and the model will read Mandarin in a Cantonese voice.
The pill on the input bar takes the name of the voice you chose, and every send from then on uses it. A voice is stored on the node: different nodes on one canvas can each use a different voice, and changing one leaves the rest alone.