I Built an A.I. Voice Assistant using PyTorch - part 1, Wake Word Detection
ฝัง
- เผยแพร่เมื่อ 7 มิ.ย. 2024
- This is a series where I walk through the engineering steps and challenges on how to build an Artificial intelligence voice assistant, similar to google home or Amazon Alexa, with Python and PyTorch on a Raspberry Pi. I leverage the latest machine and deep learning techniques to achieve this.
In this video, I show how you can build a wake word detector (keyword spotting) using recurrent neural networks specifically LSTMs.
⭐ Play and Experiment With the Latest AI Technologies at grandline.ai ⭐
Github:
github.com/LearnedVector/A-Ha...
Parts:
raspberry pi 4 model b - www.amazon.com/gp/product/B07...
ReSpeaker 2 mic array hat - www.amazon.com/gp/product/B07...
portable mini speaker - www.amazon.com/gp/product/B07...
micro sd - www.amazon.com/gp/product/B07...
First few minutes : Alright, this sounds so cool
Middle part: Da fuq
Last few minutes : Alright, this sounded cool
😂
Hahahaha
Hahaha
@@theaihacker777 is there any way u could change the voice of Google assistant to any random voice?
While internet is full of AI guru's teaching basics with some slides and a jupyter notebook, this guy actually teaches ML with a production level code. Why are you underrated !!
True.
On the learning curve, we need basic and high level AI teaching too
Amazing video! Finally someone not just showing some random jupyter notebook. I love how you show the real problems: not enough wake word samples, voice streaming, long training times, etc.
Continue if possible, I would greatly appreciate it!
You're like an alternate reality Michael Reeves
Basically a less suicidal version lol
@@soon3794 he is still yung, give him time XD
The other Micheal isn't as good with Man's AI. He just uses like google API or existing software. And Micheal Reeves just does comedy not education.
At 10:00 you said two words that sound similar, like MOLLY, and FOLLY, FALLEY. If you were to include those 2 or 3 words in your model data under a different category. Then they will never trigger when you say any of those words, except when you say hey wally.
The weather is M####F#### hot I thought that and in deep thinking this doesn't sounds right
This looks like an awesome project, can't wait to see more development!
Could you make a more detailed tutorial? I couldn't find any other videos on how to make an AI Voice Assistant, i really liked the vid altough it was sometimes hard to follow. Would really enjoy a full detailed series on this :D
The best way to learn is to mess with it my dude. Go into the GitHub and read the code, and start writing your own following his example but alter it to suit your tastes .
DUDE!!!!....You are a total BOSS!!! Thanx man. you are WAAYY better at this instructional thing than established channels/TH-camrs that (for some reason) have more subscribers, etc. Keep that stuff UP!! You are just the dude I was looking for.
Great video! High qualiy, entertaining and inspiring
This project looks super cool! I'm a little late to the party, but I think this would be awesome to revisit with the new AI chat tools! Especially something like GPT-Neo or the other open-source implementations.
thanks for the support!
Its been cool to see U
Can we make this model using raspberry Pi zero !!
the mot ovios wake word is wake up a.i.
Looking forward to next episodes.
Oh man, you are amazing! Just inspired me, thanks for the great content.
Keep it coming Michael !!! Thanks a ton!!
Your videos are amazing man. You've just earned a fan.
By the way, I'd love to see you make the speech recognition model next.
It seems that speech recognition is popular!
Been watcNice tutorialng your vids for a good few weeks now, learning new sNice tutorialt each day. my worksoftow has improved so much since watcNice tutorialng
this is insane. thank you for building this.
This video is exactly what I was looking for. All other voice assistant youtube guides use shitty Google services and other proprietary sources. Thank you. Looking forward for next vids on this topic.
Also that'd be interesting if you reveal how much time have your machine spent on all that learning.
Next video coming up soon! For wakeword, training was really fast. only spent like 30 minutes training.
@@theaihacker777 Was this process mostly fun and enjoyable? If not how did you not give up when it got hard and not get bored and frustrated? Thanks for sharing.
Great video, liking the format
Loved your explanation and flow!
honesty i never used anything like that but somehow you inspire me to start thinking how i can learn that stuff very nice video and source code thanks
This is super cool! You explained it very well, really appreciate it! :D Thanks!
“ Apple’s okay. “
Im subscribing this channel
Cool Mike. Thanks for sharing !
Ultra Cool Dude!. Fantastic instruction process and very concise. Very Good. 101% grading!.
I’m watching this video on my iPad and when you said “Hey Siri”, Siri woke up...
By the way, I’d be really interesting in seeing a video in which you explain how to set up a computer to carry out Machine Learning tasks. :)
Would love to do that!
Love u bro
I had didn't find this information even in google....thanks a lot..
I wonder why your channel is still not famous. 🤔You deserve millions of subscribers!!!
Extremely extremely intelligent AI. You asked what next to try - Can you try - Self driving car with Raspberry Pi and Pytorch. A small rover
this is so amazing!
Love this! 🤍
Finally.....I got what I was looking for.
❤️
Dude you´re awesome!!! make more of this
Awesome! I was looking for a totally offline LAN based smart A.I. like what you just presented
Objective: To control everything on my sailboat, take helm, drop anchor, play music or movies from my pi based server and work alongside my autopilot systems and chart plotter. I know often during my voyages I won't be accessing any internet but, I want to have all the same ubiquitous control as a smart home, etc. I am trying to source as much as I can from TH-cam to build a decent system.
Keep up the awesome work!
Hey, good day mate,
The project that you talked about sounds awesome. I am just checking in to know how the project is coming along and where are you headed with this project currently.
I am working on something similar on channel. Kindly check it out
Ahoy there! I'm also checking in, how's it going with the project sounds awesome and very reminiscent of what I want to do! Have you had much success?
I have the exact same idea! It's going to feel a bit like being on the enterprise.
Great video! Thank you so much!
This is very helpful! Thanks a lot. I subscribed!
You are a genius.. Subscribed !!
I really like your content and wanted to try this myself :)
Very clear explanation. Thanks a lot
Pretty interesting. For about 2 years I’ve been obsessed with turning my house into an AI assistant. Yet have the voice Overlay of ( BT from titanfall ). Yet pretty hard to do that.
Why mess with perfection?
Couldn't find part 2 and so on. Would love to do this project for myself. Thank you!
production. Thanks again!
This could be easily applied for translations, so that you can have an earplugs that instantly translates what someone is saying to you
Awesome...way to go
Thank you Mike! I'm just starting out and tNice tutorials video really helped get the basics down!
haha i looked the idea up like a year ago thinking it would be cool and everything i found was just "attatch spare phone to a speaker" lol this just showing up in my feed now is so exciting
dude, you are awesome!
Thanks for sharing. I was looking for a video that would get my head spinning then eventually put me to sleep.
"Ok, Now that we got the code.."
Wow.. that was so information dense.. SUBBED AND BELL WOW.. I'm going to learn a lot here I think :)
Amazing dude :O
This helped a lot thank you
daaaaayum good stuff homie g gangsta
Thank you very much❤️❤️❤️❤️please continue
Great learning thanks a lot.
I have gone through the video completely,if the audio signal is split with a equal diffrence for eg:2 sec then it may result in loss of information as a the split may occur at the middle of word
Great video. Have you tried the simple solution of adding examples of you saying the close phrases to the training set. Simplest way would be just to add many copies of them to the set and retrain. Possible more robust way would be to add a second stage to the process trained on only your voice.
Hey great tutorial! Looking forward to your written guide! But question/request for you: if I don't have access to expensive deep learning hardware setup like you have here, can you do another separate tutorial series on how to build/train model on say AWS/GCP?
Totally can consider that. But I think there are quite a few tutorials like that out there. Also i would recommend using google colab since it’s free and good enough for small projects
It’s a shame you didn’t keep this repo up I just got a raspberry pi and this is my intentions but I don’t know enough about ML or engineering to pull it off 😅 thanks for the content!
Interesting video. Looks like fun. You're smart af.
In other news, I've always suspected that I'm dumb af. This is now confirmed.
I'm not gonna lie when I hear your one minute speech I find some hope
It's amazing!!!
I haven’t seen the rest yet, but have you thought of creating a blacklist of all the words that sounds like Wally, and that should increase overall accuracy.
That's really great
Wow, My AI Assistant is working now Thanks!!!
Hey sorry for bothering but would it be okay if I ask you some questions on how you got it to work? Especially the raspberry Pi part
you have no idea how proud I am I even understood half of this haha
"Apple's okay" lol
thats exactly when this video got its 72nd dislike.
Why not go in the sequence you already told?
Incredible video ! I read a blog of yours on medium about LSTM, quite good explaination. Thanks man for such an awesome stuff .
Always a logical choice! but just wanted to know what the viewers are most interested in seeing next
Amazing!!!
common man, give us the next video of this series we have been waiting too long
Awesome video! Do you know if the dataset you decide to train it on has to be in .wav format? So close to getting your example working! Let me know if theres somewhere I can ask some questions regarding your code. Again great video man!
Hey Luke! If you have discord, join the discord server and I can help you there. The link is in the description.
Very informative
Hi, thanks for the great video, can you tell me how much power/cpu this wake-word implementation consumes? I've always wanted to build something like that but I'm worried about the PI having to constantly process audio to catch that wake-word that I would use 2 or 3 times a day. I was thinking that maybe for the wake-word part, some specialized hardware would be preferable, so that the PI will be under load only for a short period of time after each wake-word activation.
Hey you, I have the same computer case like you! Interesting video!
THANK YOU!
Oh yhhh......I got lost in that mathematical modelling part tho but still cool.....really cool
this is very nice
Omgurd your awesome.. First five min freaking had me laughing hard!
You make some pretty damn good content!:) When is the next video coming?
Did I just found a small creator that makes great & fun pyTorch content?
random shit!!
Love it. Your the man!
Nice making Terminator A.i
Im Currentlx using rasphy. Although its pretty neat and does EVERYTHING, it does not leave much room for the very tech savy users. I guess i will tinker a bit with your code and incorperate parts. But i need to replicate my current satellite nodes with printed pcb first. So much todo :D
Good job
Have you considered the NVIDIA Jetson Nano or the Jetson Xavier NX? Could be a better and more powerful solution than a humble pi? Looking forward to working through this project though... great videos and git. Happy New Year!
I also really like softEX, it has a very nice effect to it
Fun project.
Very nice ideea! It may be a dumb question, but here it goes: can the recorded voice be in other language than english, and by using the same principle get similar results? I'd like to create an assistant that recognizes speech input in real-time and returns information like weather, youtube videos etc.
Do you have more projects coming up? it'd be cool to see some more projects or tutorials
awesome
good job.
Thanks a lot, I just made google assistant!
Hey, what wakeword did you use
I was thinking of building too using raspberry xD. I'm surprised u didn't include cortana
You can make a video about gait recognition biometrics in python
recognized you from your walk model ????
I am currently building a bot for my thesis and want to test out everything´s working. i don´t have the parts at home and it´s kinda late to order the parts. i should be able to test it on my laptop too, right?
Well I must say that I'm very interested
Those codes make me breath deeper
this guy is genious
sir please complete full course of this AI soon...
How did you learn so much and know what to type? It just looks so complicated you are awsome
Nice bro, also Which voice in assistant didi you use?? or you just created your own??
do the NLU next!!
Bruh. Get instantly subscribed