Audio2Face with TTS Plugin -> Unknown Model

[daniel.levine]

Hi!

I have Audio2Face working on a Windows computer whether playing audio files animates the default face. Now I want to add Riva TTS so I’m following this: Audio2Face Streaming Audio Player - Riva Text to Speech Integration Example | NVIDIA On-Demand

I have a Riva Server running with TTS models on my Xavier NX DevKit hardware running the latest JP and Riva Quickstart 2.10.

However, when I tell the streaming player to play the text I type, I get an error in the console that says it doesn’t like the English-US.Female-1 (as wells as the English-US-Male-1) voice options I can select. When testing the Riva Quickstart talk.py example, I don’t specify a voice and it works. The video also doesn’t specify the voice in the UI (because it must be an older UI), and I guess uses the default and is demonstrated to work.

I want to do one more test where I use the talk.py example and specify the voice to see if I can get that to work.

Anyone have any idea what could be wrong?

[WendyGram]

Hello @daniel.levine! I’ve shared your post with the dev team for further assistance.

[daniel.levine]

Thanks! I did one more test that proved insightful. I tested using talk.py and specified the 4 different models with success:
English-US.Female-1
English-US.Male-1
English-US-RadTTS.Female-1
English-US-RadTTS.Male-1

Then I looked at the error reported by The Audio2Face app when contacting my server. It seems to be using these as the voice_name (notice the use of a - instead of a . ):
English-US-Female-1
English-US-Male-1

Is there a way for me to change the plug-in to use different strings?

[esusantolim]

Hi @daniel.levine
We are aware of the voice model issue and the fix will be included in the next A2F release.
Mean while, you can find where the extension source is for the Riva TTS sample extension via the Extension window.
Click the Folder icon and it will bring you to the folder

From there, go to these subfolders to find the tts_client.py file

and adjust line 26

[daniel.levine]

Thanks! Already found it on my own and made same fix exactly I added the RadTTS models as well.

The correct thing to do is have the extension query the Riva server (when the server address changes in the UI) for the TTS models it supports and populate the UI with those values.

Is there a node I can use for this instead of using to UI where I can pass in the UI field values and have them get submitted to the Riva Server and produce the outputs that ai can then connect to Audio2Face Core Fullface Instance’s inputs?