Azure Custom Neural Voice with Emotional Voice Styles

Question

Hi all,

Does azure custom neural voice support emotional voice styles? For example, the demo voice "Sara (Neural) - English (US)" has General, Cheerful, Angry, Sad.
Does anyone have experience with this? Would separate datasets need to be provided for each voice style, or is just one general needed?

The custom voice data docs I could find don't mention this much:
* the pricing docs[1] has prices per model, but doesn't disambiguate if the 4 voice styles above are 4 separate models or one model.
* the data guidelines[2] mention to have consistent mannerisms/speaking rate/etc, which doesn't seem to be possible with the emotional voices. Is there an updated guideline for emotional voices?

[1] https://azure.microsoft.com/en-us/pricing/details/cognitive-services/speech-services/
[2] https://learn.microsoft.com/en-us/azure/cognitive-services/speech-service/how-to-custom-voice-prepare-data

Answer 1

It looks like this link mentions voice styles is possible with custom neural voice. I guess main question remaining is whether different styles are separate models or not in terms of endpoint hosting in pricing : )

--
"Work with your voice talent to develop a "persona" that defines the overall sound and emotional tone of the custom neural voice. In the process, you'll pinpoint what "neutral" sounds like for that persona. Using the Custom Neural Voice capability, you can train a model that speaks with emotions. Define the "speaking styles" and ask your voice talent to read the script in a way that resonates the styles you want"

https://learn.microsoft.com/en-us/azure/cognitive-services/speech-service/record-custom-voice-samples#script-selection-criteria