AI accents changing - Rise 360

I am currently working on a suite of courses and, to ensure consistency across the learning experience, I am using the same AI voice, Addison 2.0, throughout. As I mentioned in my original post, I had been achieving very good results with this voice until recently.

The issue first became apparent when the voice-over for a single block in the course was unexpectedly generated with a New Zealand accent. As mentioned in my original post, I had to use a process of trial and error to eventually generate the audio with the desired Australian accent. However, after successfully achieving this and later needing to make updates to the same block, regenerating the voice-over caused it to revert back to a New Zealand accent. As a result, I was forced to switch back to the original Addison voice, which unfortunately does not provide the same level of quality as the Addison 2.0 version.

The accent change was particularly noticeable because the block contained several words that are pronounced differently in Australian and New Zealand English. This made the inconsistency quite obvious and would have been distracting for learners, resulting in an unacceptable learning experience if left unchanged.

You asked how this affects our workflow. My colleague and I typically work to tight delivery timelines, so these issues have a direct impact on our ability to meet deadlines. Omitting the voice-over is not a viable option, as it reduces accessibility and support for auditory learners. Instead, we need to spend additional time testing various workarounds that have been suggested in previous posts, hoping that one of them resolves the issue. The rework is adding additional time to complete tasks that the AI functionality should be assisting us with. 

Another challenge is inconsistent pronunciation. For example, the name of our CMS may be pronounced differently across blocks, despite using the same voice. Sometimes regenerating the audio once will resolve the issue, but in other cases multiple regenerations are required before achieving a consistent result. This adds further time and unpredictability to the development process.

One enhancement that would significantly improve this tool would be the ability to provide pronunciation guidance or feedback when generating audio. For example, allowing users to enter instructions such as "pronounce XXXX as YYYY" would provide greater control over the output and help ensure consistency across a project.

My colleague will also be leaving some feedback relating to the issues she has experienced, and the impacts it has had.