Eventually, after 6 useless videos on this topic, I found this one that clearly explains why, what and how autoencoders are doing! Great video! Bravo Valerio!
this guy is the real deal folks. i work in ML and have been doing audio for about 6 months. I wish I had found these videos back then. Top notch content here.
I am reminded of my boss asking for a 10 page report, and then asking for a 1 page summary. After some deliberation, he would again ask for expanding the summary to a 3 or 4 page report. What would happen is the important elements would get picked when we reduce and these would be elaborated when we expand again. Intuitively one catch would be the error should not be minimized to a very low value. Perhaps we would call that over fitting. The balance may be in how high is the correlation retention relative to data reduction.
This channel is gold!. It really helps to understand Ai, and its "convoluted" topics. The terminology really confuses me, but you manage to explain it in a very simple way. Watched 3 videos already ... more to go!
Well done! The first chapters weren't beneficial for me but I believe that starting with the basics really helps other people without much of a background about autoencoders in general, it's also quite kind of you to give guidance when someone doesn't know about some topic (activation functions, backpropagation, etc). Althought I can't deny that the later chapters helped clear some questions I had.
Auto-encoding and de-noising seem to be equivalent to the processes we use for selectively listening to one speaker when in an environment where multiple members are speaking simultaneously.
Thanks a lot Valerio, I really like the way you explain all of these "non-obvious" stuff ! I would really appreciate if you could make a video which deal with AEs for Anomaly detection. Do you still plan to cover this item, as you mentionned in this video ? If so, by when do you think possible to have it ready ? Thanks again, and above all : keep on going !
Thank you Gildas! I intend to cover AEs for Anomaly Detection in a self-contained video / project. Not sure when it'll come out, but it's in the pipeline ;)
@@ValerioVelardoTheSoundofAIThanks a lot Valerio! I am new to the field of audio signal processing and especially to AI approaches. Your channel helped me a lot. It was perfect to get an understanding of the whole topic. I am currently writing my master thesis in the field of anomaly detection of engine noise. I wanted to use an autoencoder as an example for the detection. I would therefore also be interested to know whether you are planning a video on this or whether it is already available on your channel. Thank you for the work and time you put into your channel! Keep up the good work! :)
@@ValerioVelardoTheSoundofAI Please conver them. I am waiting for that for over a month now and there is no other channel which cover audio processing as you 😃
working on a project that does this in some way right now :) the first thing I thought was, why use random noise when a latent vector would work so much better.
Great explanations! One question though. In the denoising example, how different is using Autoencoders from regular deep learning? I mean, you feed in the audio with the noise and test your results against the clean audio which kind of acts as the label to the data.
At the denoising part, do we always need to have same kind of noise in sound, for AE to denoise it ? For example, can samples with white noise and random noise denoised together ?
Eventually, after 6 useless videos on this topic, I found this one that clearly explains why, what and how autoencoders are doing! Great video! Bravo Valerio!
Thanks!
this guy is the real deal folks. i work in ML and have been doing audio for about 6 months. I wish I had found these videos back then. Top notch content here.
I am reminded of my boss asking for a 10 page report, and then asking for a 1 page summary. After some deliberation, he would again ask for expanding the summary to a 3 or 4 page report. What would happen is the important elements would get picked when we reduce and these would be elaborated when we expand again.
Intuitively one catch would be the error should not be minimized to a very low value. Perhaps we would call that over fitting. The balance may be in how high is the correlation retention relative to data reduction.
This channel is gold!. It really helps to understand Ai, and its "convoluted" topics. The terminology really confuses me, but you manage to explain it in a very simple way. Watched 3 videos already ... more to go!
Epic. Just what i was looking for
Valerio thanks so much for putting together such great content and providing insights into Machine Learning concepts!
Don't have any background in CS, but,2 mins into the video and you are so easy to understand.
Well done! The first chapters weren't beneficial for me but I believe that starting with the basics really helps other people without much of a background about autoencoders in general, it's also quite kind of you to give guidance when someone doesn't know about some topic (activation functions, backpropagation, etc). Althought I can't deny that the later chapters helped clear some questions I had.
Thanks for the great presentation that does not make the subject so complex and keep it as simple as possible.
27:04 yay, looking forward to watching the next video. Thanks for making this series.
Auto-encoding and de-noising seem to be equivalent to the processes we use for selectively listening to one speaker when in an environment where multiple members are speaking simultaneously.
I don't know why, but it was so easy to understand it once you made the 2D plane instend of two nodes there. Awesome
Well done. Great explanations without being oversimplified
very well explained, in perfect detail. thanks a lot man
cool! just gave me a research idea x) nice, clear and simple!
Glad I could help :)
Cool video!! I'm doing a research of speech to speech (voice convertion)
That sounds interesting!
Awesome Video! Explained really very well...
Excellent! Can't wait for next video!
Does anyone know how to find what are the important features that are stored in the latent space?
Extremely useful and easy to understand 🎉🎉🎉❤❤❤thank you so ooo much
Love all your videos! Looking forward to the next video from this series :)
Thanks!
Thanks a lot Valerio, I really like the way you explain all of these "non-obvious" stuff !
I would really appreciate if you could make a video which deal with AEs for Anomaly detection.
Do you still plan to cover this item, as you mentionned in this video ?
If so, by when do you think possible to have it ready ?
Thanks again, and above all : keep on going !
Thank you Gildas! I intend to cover AEs for Anomaly Detection in a self-contained video / project. Not sure when it'll come out, but it's in the pipeline ;)
@@ValerioVelardoTheSoundofAIThanks a lot Valerio! I am new to the field of audio signal processing and especially to AI approaches. Your channel helped me a lot. It was perfect to get an understanding of the whole topic. I am currently writing my master thesis in the field of anomaly detection of engine noise. I wanted to use an autoencoder as an example for the detection. I would therefore also be interested to know whether you are planning a video on this or whether it is already available on your channel. Thank you for the work and time you put into your channel! Keep up the good work! :)
Best video for autoencoders
Awesome!! Best explanation I have ever seen 👏👏🙌😍😍
Best explanation so far
Great video! I am keen to see the implementation of this.
What are your thoughts on using GANs to generate spectrograms?
GANs are also effective at spectrogram generation. I'll definitely cover them at some point in the future.
@@ValerioVelardoTheSoundofAI Please conver them. I am waiting for that for over a month now and there is no other channel which cover audio processing as you 😃
working on a project that does this in some way right now :) the first thing I thought was, why use random noise when a latent vector would work so much better.
I will use autoencoder for speech watermarking I hope you will aborde this subject too and thank you for this awesome videos you are the best
Great video Valerio, thank you so much!
Great explanations! One question though. In the denoising example, how different is using Autoencoders from regular deep learning? I mean, you feed in the audio with the noise and test your results against the clean audio which kind of acts as the label to the data.
I really enjoyed every secound, it was very helpful. Thank you :)
Thank you for this video. Very easy to understand ❤
please make a video on semi-supervised learning along with practical implementation.
Oh my!! Your video made a lot of sense!! thanks a lot and subscribed!!
Very Nice session. Would be grateful if you explain attention mechanism implication.
Thank you! I'll definitely cover attention / transformers in another series.
Great video! When are you going to publishing a video for anomaly detection?
which loss metric would you use for the deep convolutional auto encoder? cross entropy or RMSE still?
RMSE / MSE are a good starting point here.
Great job! Thanks for sharing😊
It's so cool how one idea like autoencoders can be used in audio, anomaly detection and I think even nlp as well! JVUVICFJ
Thanks so much!
شكراً جزيلاً 😎
Grazie mille! sei un grande
great work sir
my question is can we use kernel pca as encoder in Autoencoders because as I know it can reduce the dimensions of nonlinear data
Good explanation. Thanks!!
awesome presentation! Subscribed
Thanks!
Great video! Thank you!
At the denoising part, do we always need to have same kind of noise in sound, for AE to denoise it ? For example, can samples with white noise and random noise denoised together ?
If you have enough samples for different types of noises, you should be good to go.
can I make this project if I'm working on macbook m1 without a gpu?
Great informative video.
Thank you Nabeel!
I can't find video about anomaly detection with AEs(
Nice job 👍🏼
Thank you, so helpful
Great video 👍🏼
nice video!!! thank you so much
Thank you so much
can we implement VQ to audio classification with python?
ty :)
good
Thanks, can you cite your sources? (great explanation. thanks be)
Still waiting for the anomaly detection video :(
You still didn't explain exactly how the autoencoder works!
Valerio, can you give me an email? has some important to say/
valerio@thesoundofai.com
thank you so much