Forum Discussion
Pronunciation in AI Audio (TTS)
I am hoping someone can clarify the use of SSML with Eleven Labs voices in Storyline 360. I've read variously on E-Learning Heroes that SSML is not supported with Eleven Labs voices or that it is supported but only for <phoneme> and <break> or it's supported for other tags too. I've been trying to use it with the phoneme tag but when I generate speech, it simply skips over the tagged content.
This is a sample of one use I'm trying to get to work:
Repair of this injury can be addressed laparoscopically in stable patients. The first priority is adequate exposure. Additional ports should be placed as needed to isolate the injury and improve visualization. If the bowel is friable it can be handled indirectly using <speak><phoneme alphabet="ipa" ph="æ.tɹ.ə.mˈæɾɪk">atraumatic</phoneme></speak> graspers on the mesentery, and the edges of an enterotomy can be approximated to limit contamination if the injury is small.
When generated, all the speech between the opening <speak> and close </speak> tag is simply skipped but the rest of the speech is rendered. Since I am not sure if SSML is even supported, I'm not sure if I'm not rendering the tags correctly, or if it simply isn't supported.
Hello JohnDwyer-f5da1,
I'm glad to see that you're exploring the use of SSML Tags in your Storyline course! I can confirm that AI Assistant offers limited SSML support because its voices are built to interpret word context and adjust delivery naturally. That will contribute to the issue you are experiencing.
- What voice are you currently using? You'll want to confirm that SSML tags are supported for the selected voice.
I suggest using the other available Text-To-Speech option in Storyline, located under Insert> Audio> Text to Speech, so you have more control over pronunciation.
To test this on my end, I simply moved the <speak> </speak> tags to the beginning and end of the sample text you provided instead of with the <phoneme> tags.
Here's a quick visual of my experience in Storyline as well as the modified tags in your sample text.
<speak>Repair of this injury can be addressed laparoscopically in stable patients. The first priority is adequate exposure. Additional ports should be placed as needed to isolate the injury and improve visualization. If the bowel is friable it can be handled indirectly using <phoneme alphabet="ipa" ph="æ.tɹ.ə.mˈæɾɪk">atraumatic</phoneme>graspers on the mesentery, and the edges of an enterotomy can be approximated to limit contamination if the injury is small.</speak>
I hope this helps! Please let me know if you have any additional questions!
7 Replies
Hello JohnDwyer-f5da1,
I'm glad to see that you're exploring the use of SSML Tags in your Storyline course! I can confirm that AI Assistant offers limited SSML support because its voices are built to interpret word context and adjust delivery naturally. That will contribute to the issue you are experiencing.
- What voice are you currently using? You'll want to confirm that SSML tags are supported for the selected voice.
I suggest using the other available Text-To-Speech option in Storyline, located under Insert> Audio> Text to Speech, so you have more control over pronunciation.
To test this on my end, I simply moved the <speak> </speak> tags to the beginning and end of the sample text you provided instead of with the <phoneme> tags.
Here's a quick visual of my experience in Storyline as well as the modified tags in your sample text.
<speak>Repair of this injury can be addressed laparoscopically in stable patients. The first priority is adequate exposure. Additional ports should be placed as needed to isolate the injury and improve visualization. If the bowel is friable it can be handled indirectly using <phoneme alphabet="ipa" ph="æ.tɹ.ə.mˈæɾɪk">atraumatic</phoneme>graspers on the mesentery, and the edges of an enterotomy can be approximated to limit contamination if the injury is small.</speak>
I hope this helps! Please let me know if you have any additional questions!
Hi Luciana,
The way you have it set up now is how I originally had it set up and that didn't work for me either. I'm using Matilda. It looks like the tags aren't supported for that voice.
The other TTS sounds too robotic for my client.
Totally agree. We use a lot of scientific Latin-based terminology in our training, and it can take me hours to get it to say the word the right way (it may say it right in one paragraph, but say it wrong in another, even if spelled exactly the same). The standard TTS sounds horrible - definitely not a solution for our trainings.
Where are you getting the IPA alphabet from? Is there a website one can access?
Hi there!
There are several websites I've worked from. I like
https://ipa.typeit.org/full/ for typing help.
I also find
https://www.ipachart.com/ to be really helpful.
Thank you!
I have not done much with SSML, we have been using the TTS voices so far, but the AI voices are generally so much better!
With the normal voices, I have my tricks how to get them to pronounce our company name and brand names correctly (or close enough), but that does not work for the AI voices. The main issue is that a short text snippet (for testing the voice) may be pronounced correctly, if I paste the entire text, it doesn't. Or, as mentioned above, pronunciation varies through the text and the whole project. Not acceptable I think.
This being an AI voice, I wonder if there is (or could be implemented) a way to use a kind of "master prompt" or rules of pronunciation for certain words in certain languages, similar to a glossary for translations. That way, you could have a file with the settings for a specific client, language, whatever, similar to a colour or font theme, which you just read into storyline (or of late even Rise, albeit still in labs I think) and you are good to go.I work for the same company all the time, but especially for those who have several clients, this could be a massive time-saver.
