Unable to start llamaspeak (AGX Orin 64GB)

First off, I admit I am still new to Orin, but I am pretty good at following instructions I think!

However, I get this error when I fire up the llama_speak example in the browser console:

“Firefox can’t establish a connection to the server at wss://192.168.1.252:49000/”

Also this:
TypeError: input is undefined

on this line (13) in audioWorkers.js:

var samples = new Int16Array(input.length);

It’s probably a silly error, and it’s probably my fault. But I’d love some pointers if anyone has seen this!

More context: Riva is started successfully and running on the default port. I am able to successfully do TTS and STT using the running instance.

@guitargice can you try chrome/chromium instead?

I got further on Chromium, thanks Dusty! Firefox didn’t like the self-signed SSL Cert so it didn’t open the websocket connection. Now the TTS / STT isn’t working, but I’m troubleshooting that. We can resolve this one. Thanks again!

Ok thanks @guitargice, I need to document that. There are a couple --debug options like --web-trace if I recall correctly if you aren’t sure if the audio is being received from the browser, ect. That kind of thing is why I always test the ASR/TTS in stages separately first before the web agent, but can still sometimes be tricky.

Gotcha, thanks @dusty_nv ! I tested ASR and TTS successfully. The direct TTS audio output was super choppy, but when written to a file, the file plays cleanly with aplay. I did notice that the “Select voice” dropdown in the llamaspeak ui was not populated with voices. I haven’t dug deeper into the code to see if that’s hardcoded or retrieved via API somewhere, but I’ll let you know what my spelunking turns up if relevant. Thanks again for your help.

RE: the voices, that may be because I switched over to using --tts=xtts while RIVA team resolves this streaming TTS issue, and with XTTS I was able to enumerate the voice list from the model and auto-populate it in the web UI…but have not changed it over in the Riva TTS plugin yet (because of the streaming TTS issue and not currently using it for TTS)

I think it generates audio to the wav file cleanly, because that is using offline TTS mode (i.e. returns the entire audio buffer at once), as opposed to streaming TTS (which this issue seems particular to)