Forum Discussion
AI Text to Speech V3 Hallucinations
ETA: I have not received any notifications for updates as has been suggested in another posted by AshleyBlain-ad7 21 days ago.
Hi DonWallbaum,
Thanks for reaching out and sharing your experience with AI text-to-speech!
The v3 model is expressive and offers a wide emotional range. For best results, you might want to try these prompting techniques. If you’re looking for more consistent output, we suggest switching to the Multilingual v2 model for now.
I also checked the discussion you mentioned about missing Storyline update notifications. Let’s start by confirming your current version.
- Open Storyline
- Go to Help
- Select About Articulate Storyline
Please let us know what version you see!
Thank you, Kelly. It appears I have: v3.112.36406.0
Now, while I have you... and again, thank you for your time on this, working with another Text-to-Speech tool (Murf) has better results if a smaller amount of text is entered per block, then the individual blocks are melded together to form the final output.
I see that Storyline 360 has a limit of 3000 characters per slide. but I see no way to create more text blocks, of say one sentence per block, and then having Storyline meld them into a multi-sentence audio file. And I don't even know if such an approach would work better in preventing hallucinations than having all the slide text in one block but within the 3000 character limit.
Do you have any thoughts on having a larger number of small text segments - one sentence each - that then meld into a larger audio file: is this even possible in StoryLine and/or would that approach have any merit?
Again, my thanks as I work through my Articulate confusions.Don
Hi DonWallbaum,
Great to see that Kelly has been helping you. Welcome to authoring in Articulate 360!
First, I'd like to get you updated to the current version of Storyline, which is Build 3.114.36655.0. If you're not seeing the update alert at the top of the desktop launcher, you can also manually check for updates. Click the downward arrow next to your profile icon, then select Check for Updates.
To your follow-up question, Storyline currently does not support a built-in way to input multiple text blocks and automatically merge them into a single continuous audio file within the Text-to-Speech editor. Each TTS object is generated independently based on the text provided.
As a workaround, you can generate separate audio clips for smaller segments of text (e.g. sentence by sentence). Then insert them as individual audio files and sequence them on the Timeline to play back seamlessly. This can give you more control and may help reduce inconsistencies. However, it’s not guaranteed that this approach will eliminate all hallucinations. If consistency is the top priority, sticking with the Multilingual v2 model is still the most reliable option that we suggest.
Let me know how it goes!
Thank you, Steven.
Being a newbie to Articulate, I don't have the body of knowledge to comment competently on issues I am having with it, though to date, the hallucination is really the only irksome element I have found.
More generally.... one reason the group I am with is moving away from Captivate is it released a version that was incompatible with its old version, with no capability at the time for the "new" Captivate to work with "classic" Captivate files, and there was no translator to take the old "classic" project and make them compatible with the new.
I bring this up only because I see the same mindset here in my brief experience with Articulate.... that is promised functionality (AI V3 TTS) that simply is not ready for prime time, and the paying user base are the beta testers.
Cory Doctorow came up for a term for this degradation of services back in 2022... sadly, his analysis is more accurate than I would have wished.
But, I digress. Thank youall here for the support you have offered this newbie! The generous nature of your time and experience I greatly appreciate!
Don
