This message was deleted.
# helpdesk
s
This message was deleted.
m
cc @boundless-energy-78552
b
Hey, what do you mean by the audio recognition rate?
The GPTTrack is the downstream track, this is where Kitt “speaks”
Since we’re not reencoding the audio on the server side, you need to edit the client parameters to edit the sample rate
g
When I started to talk to the ai, the subtitles displayed were a little different from what I said. I wanted to improve the recognition effect of this part. How should I adjust it?
m
@boundless-energy-78552 I think audio recognition is the sample rate passed into Google STT.
g
yep
b
The code is here, I don’t think GCP gives options to edit the interim results
The options are defined here
(We’re just forwarding the Opus data from the client to GCP)
g
Are the parameters here the result of your tuning?
Copy code
config := &sttpb.RecognitionConfig{
		Model:             "command_and_search",
		UseEnhanced:       true,
		ProfanityFilter:   true, // replacing all but the initial character in each filtered word with asterisks, e.g. "f***".
		Encoding:          sttpb.RecognitionConfig_OGG_OPUS,
		SampleRateHertz:   int32(t.rtpCodec.ClockRate),
		AudioChannelCount: int32(t.rtpCodec.Channels),
		LanguageCode:      t.language.TranscriberCode,}
👍 1
Your official website demo STT recognition effect is very good, I want that effect
b
This is the same code, so this is weird, do you have latency?
g
Does this have anything to do with the server? Our front-end code is deployed in vercel and the server code is deployed in Azure.
🤔 2
b
I’m not sure, are you using LiveKit Cloud?
g
yep
Can I provide you with a link so you can test it?
b
Yes sure
g
You can ask some questions and I will have a record of the conversation so I can check whether it is accurate or not. https://kitt.lzl.dev/rooms/blkv-p02w
b
It doesn’t seem to work at all for me nvm my airpods micro is broken
My english accent isn’t perfect, I tested in french and it is working fine. The AI is answering me in English, I guess you made some changes here?
g
yep, we're tweaking our ai model at the bottom
b
this is really cool btw
g
I got a transcript of your conversation. I'll send it to you. Is it accurate?
👍 1
b
I don’t think the English one will be accurate, but the french one will
g
pourquoi tu parles anglais oui je suis encore là merci j'en ai aucun aucune idée je sais juste un développeur
b
Yep this is accurate
maybe just
je sais
to
je suis
. But it was mb my fault
g
I guess it's my end of the network.
b
Maybe @magnificent-art-43333 can try to speak in English
m
I can give it a try a bit later today or tomorrow — plane about to take off 🛫
✈️ 1
g
Well, thank you for your patience. I really appreciate it.