Forum Discussion
The voice changes after using a
hi there, I am using the text to speech to generate audio with Matthew US neural voice.
SSML and the fine-tuning of the closed captions are a very good progress in terms of features and customisation options.
A general issue I find is that when using the following SSML tag the voice changes. Example
<say-as interpret-as="characters">ABC</say-as>
Matthew's voice changes into a different male voice.
Is this normal behaviour?
Thanks
Bobby
13 Replies
Hi Bobby,
Thanks for reaching out!
I tested the behavior using unmodified text, and text that has been modified by the <say-as> tag and I was unable to observe any changes in the voice of the resulting output. Here's a screenshot of the text that I used for testing:
We're also not seeing this behavior reported by other users. Since the behavior appears isolated, try repairing your Storyline 360 installation. If the issue persists, please share a copy of your project file here or in private by opening a support case so we can test to see if we'll get similar results when testing using the same file and text. I've included the file I used for testing and the sample published output for your reference.
Hi there Jose and thanks for the quick reply. I used some of the tags from my original file in your file and all was good.
Then pasted longer parts and got the same issue. Quite interesting. I will do what you suggested. Just that I have to go through our IT and that would take time.
Thanks again
Bobby
I'm seeing a similar issue. I've been experiencing issues where some other male voices are changing and they end up sounding exactly like the Matthew voice. In some cases, the audio will start out with the selected voice, then suddenly change to the Matthew voice. I've seen this both with and without SSML tags.
Hi L_Hendrickson,
Happy to help! What you have reported is strange, as the selected voice suddenly changes while the course is playing. I'd like to start with the troubleshooting steps below.
If you're already working on your local drive, try running these steps to fix possible installation issues.
We'd be glad to investigate further if the problem persists. You may share your Storyline project here or privately in a support case, where our Support Engineers will work with you one-on-one. We'll delete it when we're done testing.
Just experienced this with the Neural Voice Mia (Spanish-Mexico). One sentence in the middle of a longer clip changed to a different female voice. I tested with Lupe (Spanish-Latin America) and it happened again. Oddly, the male voice Andrés (Spanish-Mexico) did not change.
After trying a bunch of stuff to fix it, I created a blank Storyline file and added the sentence word by word and was able to narrow the issue down to the very end of the sentence. Our translation company sent us a sentence that ended with "y la ID de mesa" ("and the table ID"). When that got added in, the voice changed. Luckily I was able to resolve this by just changing the end of the sentence to "y el numero de la mesa" ("and the table number").
You can test this by choosing Mia and inserting the following: "Quiero el numero de mesa. Quiero ID de mesa. Quiero el numero de mesa." The second sentence changes, with the speaker pronouncing ID as "eed." Hope this helps!
Any updates on this issue? I'm experiencing this now. Matthew's voice changes in the middle of a pretty short string of text. No SSML code; just typing the text.
Hello TamaraCraft-62f,
Happy to check this for you!
Would you mind sharing a copy of your project file here or privately by opening a support case so we can examine the script and the audio file that was generated? And does the problem happen if you switch to another voice, or does this only occur when Matthew is selected?
Hello Jose,
I will open a support case.
Thank you!
I'm experiencing the same issue with Neural Amy UK - the voice up until the tag <say-as interpret-as="characters">[text]</say-as> sounds differently (like Standard Amy UK, more robotic). Once I remove the tag, the voice fixes itself and it is Neural Amy UK throughout.
I decided to check the Amazon Polly developers guide - since the Storyline TTS voices are Amazon Polly voices - and this is what it says on page 247 in the "Controlling how special types of words are spoken" section: "The <say-as> tag is supported by generative, long-form, neural, and standard TTS engines.
Note, however, that if Amazon Polly is using a neural voice and encounters the <say-as> tag with
the characters option at runtime, the affected sentence will be synthesized using the related
standard voice."So, in other words, the voices are intentionally changing for the <say-as interpret-as="characters"> tag - and there is nothing Articulate can do about it.
Thanks! Follow up.... I'm not using SSML and I do not have a <say-as> tag in the text-to-speech. Is there some combination of words or punctuation that might be interpreted as a <say-as> tag by Storyline? Something I can avoid using in the future. There are just words, periods, and commas in the text.
Hi TamaraCraft-62f,
There isn't a combination of text that Storyline 360 will interpret as SSML tags. You need the <speak> tags to tell Storyline 360 you're using tags for the text. Are you still encountering issues with the voices used by your text-to-speech audio? If yes, would you mind sharing a copy of your project file here so we can check the script used to generate the audio for any issues?
