[Bug]: 3D Blendshape Data Not Generating for Super Realistic Voices

[jillbourque]

What happened?

The 3D Blendshape SSML function for the Super Realistic Voices seems to be broken. It was working until about a month ago. When we send an SSML request we get audio but no 3D Blendshape data for these voices:

"Super Realistic" Voices NOT Sending 3D Blendshapes en-US-EmmaNeural - NOT Sending TTS 3D Blendshape Data en-US-AndrewNeural - NOT Sending TTS 3D Blendshape Data

Yet, these "regular" voices are still correctly sending blendshape data:

These "Regular" Voices ARE sending 3D Blendshapes en-US-JennyNeural - Sending TTS 3D Blendshape Data Correctly en-US-NancyNeural - Sending TTS 3D Blendshape Data Correctly en-US-SteffanNeural - Sending TTS 3D Blendshape Data Correctly

Blog on Super Realistic Voices https://techcommunity.microsoft.com/t5/ai-azure-ai-services-blog/introducing-super-realistic-ai-voices-optimized-for/ba-p/3933744

Version

1.36.0 (Latest)

What browser/platform are you seeing the problem on?

Chrome

Relevant log output

No response

[yulin-li]

Hi, the blendshape support for these voices is being added. We will update this issue when it goes live.

[peterpanseu]

Hi @jillbourque, the problem has been fixed and now the realistic voices can support 3D Blendshape. Please have a try.

[jillbourque]

Hi @jillbourque, the problem has been fixed and now the realistic voices can support 3D Blendshape. Please have a try.

Hi @peterpanseu and @yulin-li

I tested and Emma is now working but Andrew is still not working. And to clarify, this voice was sending blendshape data just fine until about a month ago.

I have tested with the following results:

NOT Sending TTS 3D Blendshape Data en-US-AndrewNeural

Sending TTS 3D Blendshape Data Correctly en-US-AndrewMultilingualNeural en-US-JennyNeural en-US-NancyNeural en-US-SteffanNeural en-US-EmmaNeural en-US-EmmaMultilingualNeural en-US-AvaNeural en-US-AvaMultilingualNeural en-US-BrianNeural en-US-BrianMultilingualNeural en-US-GuyNeural en-US-AriaNeural en-US-DavisNeural en-US-JaneNeural en-US-JasonNeural en-US-SaraNeural en-US-TonyNeural en-US-AmberNeural en-US-AnaNeural en-US-AshleyNeural
en-US-BrandonNeural en-US-AIGenerate1Neural

[peterpanseu]

@jillbourque could you please try a different text? It could be that the text you used for testing AndrewNeural has been used previously and the problematic result has been cached. By using a new text, it will be a new synthesis using the newly deployed voice. In the test on our side, AndrewNeural can return BlendShape data. Please have a try.

[jillbourque]

Hi @peterpanseu Ah, yes! Now it is working correctly.

[jillbourque]

Thank you!

[peterpanseu]

My pleasure.

[Richa10729837]

My pleasure.

@peterpanseu thank you for support. @jillbourque I appreciate your cooperation!!