AI text-to-speech voice inconsistencies?

Hi everyone,

I'm working on a Storyline project that includes a customer service interaction. For one of the characters, I'm using the Zoe voice from the voice library (described as charismatic, gritty, and engaging). However, when using the v3 voice model, the voice intermittently shifts between the expected delivery and what sounds like a completely different voice with a strong Southern U.S. accent and noticeably different cadence.

Occasionally, switching to v2.5 and then back to v3 temporarily resolves the issue, but the results are inconsistent. As a result, generating audio has become quite time-consuming and frustrating.

I'm curious whether this is a known issue with the v3 model and, if so, whether there is a workaround or planned fix. Additionally, are there other African American female voices available within Storyline that are considered stable, long-term options and less likely to be retired from the platform? The options seem to be quite limited from what I've seen.

Any insights or recommendations would be greatly appreciated. Thanks in advance!