The character responses using Eleven Labs voices seems to have an issue
User input
Saying “Hi how are you” to the character.
What happened?
Hi! From last night if I select an Eleven Labs voice for the character their responses are most of the time taking longer than usual and then when do finally speak their response is broken up, like they will say part of a word or sentence, then there is brief silence, then they will say the rest of the word, then brief silence again, etc, for the whole response.
This broken up speech seems to only happen if their response is slow. Also I have noticed that if I do an InvokeSpeech() in code to get them to say something specific, then their response is fast and unbroken.
My internet is fine. I can make a video of this if necessary.
What did you expect?
For the character to respond quickly without broken up speech.
Steps to reproduce
Speak to the character.
Redacted logs or response
No error messages.
I checked the relevant documentation, resources, and similar forum posts before posting.
on
I confirm that I did not include API keys, tokens, passwords, secrets, private character data, private conversation logs, or customer data.
Hi K3, thanks, using Gemini 3.5 fixed it. I’ve no problem using that permanently for all of them, it does a very good job and is very quick, but 1 character I have, a cyborg, works really well with Gemma 4 31B Fast which is real time I think, so it would be great at some point if just that 1 character can use that again. I will switch the one’s using real time LLM’s over to Gemini 3.5, thanks for your help
EDIT : I spoke too soon, changing LLM didn’t fix it unfortunately.
No in my very quick test in the Playground, it’s happening again in my app though unfortunately, it seems to come and go, its been happening for a while but rarely but since last night it has persisted. I made some edits to the ConvaiNPCAudioManager script last year some time and it was working fine 99% of the time to quicken up responses but have reverted them and tested and its still the same.
So I did some digging on this, I saved the raw audio my app receives from the server into .wav files on the Quest 3 so I could listen to what comes in and I also had a log file timestamp what time the wav files got to my Quest 3.
There’s loads of little clips and each one sounds ok on it’s own, no broken audio in the wav files themselves, but when I looked at the timings the audio is arriving, its arriving in bursts of between 2 and 10 seconds between parts of a response, so say the avatar was saying “I am fine thanks, how are you”, the audio wav file parts received might say “I am fine” at 0 seconds, then then " thanks, how" received at 4 seconds, then " are you" at 8 seconds.
So it looks like my app is sat there with nothing to play because of these gaps in the audio.
When the server response comes back quickly its fine, it sounds ok, but for some reason most of the time its sending chunks of the response with sometimes several second gaps in between. I don’t understand though that when I use InvokeSpeech() the character response is pretty sure fine, maybe that doesn’t code doesn’t chunk the response?
I could be entirely wrong and I am ready to be told it is all my fault because it’s entirely possible because I edited ConvaiNPCAudioManager last year but its been working ok for 9 or 10 months now, with the occasional hickup.
