How to build a real-time AI assistant (with voice and vision)
ฝัง
- เผยแพร่เมื่อ 7 ส.ค. 2024
- This is a new version of my AI assistant, this time using LiveKit (livekit.io.) This is the same platform OpenAI used to build their ChatGPT assistant.
The source code of my example is here: github.com/svpino/livekit-ass....
I teach a live, interactive program that'll help you build production-ready Machine Learning systems from the ground up. Check it out here:
www.ml.school
To keep up with my content:
• Twitter/X: / svpino
• LinkedIn: / svpino
🔔 Subscribe for more stories: / @underfitted - วิทยาศาสตร์และเทคโนโลยี
Incredible video. You're taking your content to the next level. Keep up the good work and thankyou for all you do.
Sir, your AI voice assistant demo's are one of the most valuable and appreciated youtube videos I have come across. Please keep them coming and also would be great to do a demo with Groq for solving latency issue. You are doing great work man and your students are really appriciating it! Thanks a lot brother!
Great video. Keep these coming.
Really cool stuff... I def plan to recreate some of these things along with you when I have a bit more time at my computer. Just a thought, adding screen capture in with this would be pretty cool too to get help with whatever applications you're in... I would imagine you could include both camera and screenshot images in the same context and it should be able to distinguish which you're asking about.. or build a different tool that it can function call for that. Can't wait until we get some slightly more expressive voices as an option like OpenAI teased us with.
You turned your videos to the next level which is pretty amazing. I would like you to ask if is possible, show us how to use open source models in those apps. Thank you in advance.
Amazing content with solid explanation. Thanks very much 👏
Glad you liked it!
from livekit import agents, rtc
ImportError: cannot import name 'agents' from 'livekit' (unknown location). I am getting this error.
very interesting! thank you!
Thank you, It is very interesting.
This was a really cool stuff. Thanks for sharing such a quality stuff. I ran it on vscode and it worked. The main problem is the latency. It took like 10s for the conversation. Not sure if it is because of the free account or their websocket API has the issue.
Fantastic, thank you very much. How would you adapt your code if you have multiple participants?
obrigado, fantástico o vídeo 👏👏👏👏👏👏👏👏👏
Incredible!. I tried the same thing there is a noticeable delay. I am not sure if it is the TTS or STT or the LLM API itself. While yours and the default demo at Livekit is damn fast.
Great video, keep going. Just wonder if you can add barge-in function?
Hi, this is great. I tried and its working. I tried to introduce my self, then I ask, "Do you know my name ?" the response is "I'm not able to remember past conversations." So, Can we make it remember the conversations, and also "remember" my face ?
That actually a good question. I saw a paper published about persistent memory: short term memory, medium long term memory and building an automated RAG system to automatically RAG information to long term memory, so we can create an assistant that’s goal orientated and proactively manages you and towards meeting your goals.
Hi, I am working on creating a closed lan-network, using per to per, and will input a live AI agent, locally stored, getting knowledge from LLM, and wonder if it is possible to have this kind of system then running without using internet?
Can be used with google flash? Thanks for super content!❤
great video as always. Does this system keep chat history?
This is really cool. As someone with python knowledge how would you suggest I get started?
Any roadmaps or videos you published?
Great info, thx! Is there a way to use local LLM (like ollama, local AI etc) on this platform instead of openai?
Nice men thanks
I hope we can we use gemini 1.5 pro? I will try to make this changes in old code
This is great!
Hi ..thanks for the nice video.. Can we use local LLM (like ollama) instead of a paid one..?
sir can you help me why my assistant isnt talking back or nothings happening but its recognising in chat what im saying
Hi, great video! How to change the source code that it captures my screen, desktop. Thanks.
Got stuck with the API Keys, i'm not sure how to set them
Hi , i start test with all your steps and got stuck on the fact that text and audio from the openai api are not transferred to livekit, all requests pass in the terminal , tried many solutions...doesnt work
I think I may have a similar issue on Windows there is some problem with the ffmpeg library.
I wish the code would process the timeline.
so i cannaot code can you make toturial for using ph3 which is free and have vision and also use visper ai to convert text to speech and other free tools so minimizing the cost to completely zero I am a student trying out these stuff and don't wanna pay or don't have money to pay for the API or other things so please make a toturial using all the free and open source tools
Hola Santiago saludos desde Cuba, acabo de ver en su perfil de Github que es un coterráneo. Su contenido es muy valioso, por aquí un ingeniero de IA tratando de salir adelante en este mundo. Me encantaría poder contactar con usted y ayudarle en algún proyecto.
would love to see how to to implement Deepgram TTS instead of OPenAI !
It’s actually very simple: simpler than what I had to do to get OpenAI working
@@underfitted Hey thank for replying! This would be an awesome! Also using maybe openrouter as well! Wild ideas come to mind!
Santiago, the voices of assistant are only in english? or also in spanish? abrazo crack!
They speak Spanish too
could i get this on my smartphone?
As is, no. You’ll need to rewrite it in a phone-friendly language
Is this functionally any different than your previous video?
While they work the same for the demo, my previous code is very brittle. This one is much better because I’m using an entire existing infrastructure to support it.
your code generally doesn't run in VScode in windows... some things seem platform dependent unfortunately
I don’t think it’s a problem with my code… it’s a problem with Windows. Try WSL.
Curious about the latency. I noticed that you cut the video after each question (after 19:55), so I am assuming it was a few seconds?
It wasn’t bad, but GPT-4o is not as fast as it could be, so you definitely have to wait a second or so for an answer
@@underfitted , would be nice to add at least one answer in real time so we would get more realistic picture of it.
Dangggggggg!!!
now i keep running into troubles when using this code,
I would love to be able to discuss this so i can get it fixed i want to implement some features to see if it can work for something else to.
Yes the API keys do not pass -- what are you experiencing?
@@AI_by_AI_007 Hi Yes,
So i had to rework the code a little bit to get everything working again.
And now its working great except that the voice of the AI is not working and i cannot give voice commands anymore.
@@aaronwenniger7966How do you have do for the livekit api key ?
How for the livekit api key ?
@@Noahperaudon ?
2:38
Has anyone gotten this to work on Windows?
No I've got an error at line 12
from livekit.agents.voice_assistant import AssistantContext, VoiceAssistant
ImportError: cannot import name 'AssistantContext' from 'livekit.agents.voice_assistant' (D:\Github\livekit-assistant\.venv\Lib\site-packages\livekit\agents\voice_assistant\__init__.py)
@@LesBrickodeurs yes the SDKs have changed and a lot of this no longer works off the bat ..im still playing around but it probably needs redoing
Hey I have a issue with key api livekit its telling me error like its invalid
Me as well -- YOU on windows or MAC as you try this?
@@AI_by_AI_007 windows
@@AI_by_AI_007Windows
@@AI_by_AI_007 How can I access to the API Keys? could you please tell me step by step.
Yeah, this is cool but not helping. I can open the LiveKit interface but can't find a way to get the agent to connect. API keys all correct
I think the track management piece has changed since you made this. Running into no local_participant attribute for the Room object
Shame that the hook with videos is start open source and then get people draw into handling supporting functions via a commercial platform (for $$$)