Encontre sua tribo de voz

Nothing in my config changed, but one file in the middle of a batch came back with a full British accent. One word even drifted Scottish. The files before and after it are perfectly American. Same voice setting, same text style, same day.

I'm narrating a full book chapter and one segment came back with the same sentence spoken twice in a row. Not stuttered — just cleanly repeated, like a stuck record. A second file did it worse: it looped an entire paragraph until the audio ran out.

About one file in twenty comes back as complete gibberish. Not an accent thing — slurred, mumbled phonemes that don't form any real words. Here's what makes it sneaky: the sentences around it are perfectly fine. A spot-check would sail right past it, and you'd only catch it if you listened to the whole file.

Every few files, the audio just stops a word short — like the narrator got cut off before finishing the sentence. No error, nothing in the logs, and the API even reports success. It's not random either. It happens most on my longest paragraphs, and I've noticed a pattern with sentences that end in a period. The final word just... disappears.