Forum Discussion

Paul_Atleos's avatar
Paul_Atleos
Community Member
8 days ago

Storyline 360 AI Voices V3

Hi,

I'm curious about other people's experience with AI voices now that V3 is the default.

Personally, I'm not impressed... they seem quite buggy.  I've found that v3 voices take a very long time to generate when compared to v2, often seem to insert random glitchy noises, and sometimes just can't even be played back.

Further, I've noticed that the voices themselves, when they do work, seem to have lost their original character. I regularly use Lynda and Lewis, 2 Scottish voices.  These were great in v2, but with v3 Lynda now sounds American and Lewis seems to have become Australian.

In short, I'll be sticking with the V2 voices, but just wondering if my experience is unusual or if others are finding the same.

11 Replies

  • I just tried the v3 voices and experienced SIGNIFICANT hallucinations. Where should I report this to Articulate? 

      • StevenBenassi's avatar
        StevenBenassi
        Staff

        Hi Paul_Atleos​,

        Sorry to hear you're having trouble with the V3 model of AI voices in Storyline.

        I see that you've been working with my teammate Robby, through a support case. As he shared, we have this behavior logged as a bug in AI Assistant. This bug causes different accents to be generated when using the V3 model of AI Audio. We're seeing this bug in both Storyline and Rise.

        We've included your voice in the bug report so we can notify you as soon as a fix is ready. As a workaround, we suggest using the V2 model, where this issue is not observed.

        I'm sorry if this has been slowing you down!

  • Paul_Atleos's avatar
    Paul_Atleos
    Community Member

    I was beginning to wonder if it was just me... in a way I'm glad to hear that other people are having the same issues.

    I think v2 of the voices is great - I very rarely get issues with hallucinations or glitch noises and on the whole am loving AI TTS.   But with v3 it's a pretty regular occurrence: to the point where it's pretty much unusable, as I spend more time editing out the random noises it inserts than I do generating the original audio.

  • PlusPortCT's avatar
    PlusPortCT
    Community Member

    I’m also running into quite a few issues with v3. Sometimes, the voice feature hallucinates like crazy.. I have no idea where it’s getting these stories from! I've heard glitchy noises aswel. Also, there's a weird bug: when a short page title is followed by a line break, the start of the next sentence just disappears.

    • StevenBenassi's avatar
      StevenBenassi
      Staff

      Hi PlusPortCT​,

      Sorry to hear you're also experiencing erratic behavior with AI Voices!

      The hallucinations are related to a bug we're currently tracking for AI Assistant. This bug causes different accents to be generated when using the V3 AI Audio model in Rise and Storyline. I've linked you to the bug report so we can notify you as soon as a fix is available. As a workaround, please use the V2 model where this issue cannot be observed.

      Also, do you mind providing an example of the line break issue you've shared? Seeing a screen recording of the behavior, and a sample .story file can help us troubleshoot this better. Feel free to upload those items here in the discussion or privately through a support case.

  • TracyBritan's avatar
    TracyBritan
    Community Member

    So far, I've enjoyed the v3 AI voices. Especially since they pronounce my company's name correctly, which some of the V2 compatible voices did not, and I regularly had to change the spelling to get the voice to say it correctly. Anyway, I was wondering if Articulate will be doing a training on tips for prompting or tagging text-to-speech to specify inflection, etc. For example, I can't figure out how to de-emphasize the word "not" in a comparison sentence. I'm using "Ginger" and she keeps saying "Confidence is NOT accuracy." I want it to be more like CONfidence is not ACCURACY."

    • Paul_Atleos's avatar
      Paul_Atleos
      Community Member

      I created a post on the topic of emphasis in TTS a few months ago (AI Voice Generation emphasis in SL | E-Learning Heroes)...  a few people provided tips but there's no official guidance from Articulate and no sign of any coming soon.

      One tip was that using speech marks and capitals to add emphasis can sometimes give good results.  So in your example:
      "Confidence" is not "Accuracy".

      This still gives patchy results though, so often all you can do is create short snippets with the statement you want to emphasize, and just keep regenerating until you get the result you want.