Part 3-Model Creation-Audio Classification Project Using Deep Learning

[@krishnaik06]

The output that i got is wrong.. So will make this model better ussing CNN

[@Kirikiri085]

Sir what are applications of this and where we used in real life?

[@Kirikiri085]

I'm having this project tomorrow

[@K-mk6pc]

Sir Thanks for making these videos free Sir love you

[@sirshadutta2253]

Sir can you please make a video on what are the minimum skills to get a job or internship as an ML engineer

[@santhosh20071993]

DropOut of .5 have to be used with caution. Basically we throw 50% of training features to a dustbin and same effect propagates to hidden layers

[@navronaman]

You are amazing sir

[@aadityaparthasarathy4743]

Just a doubt. How do we listen to microphone in real time and then use that as input to a model? I want to make a web app out of it

[@chalakajayasinghe]

Have you done this?

[@pavithramuniraj8128]

Hi there, have you completed this? I need to do similar thing.

[@balapranav5364]

Please make video on Gower distance used in clustering

[@romyullislam5671]

model.predict_classes(prediction_feature) here predict_classes() function is not working

[@pazussajuan]

yeah, you can use model.predict and you obtain the probability vector of prediction

[@bhanuyaswanthreddy300]

Sir can you please tell whats the ml algorithm you used in this audio classification?

[@gregoiremafutala9897]

Thank you Krish Naik for this video series that will help me in my research, it's a good job ;) 04:24 Given that we can rerun the model training several times and keep the updates in order to increase the accuracy of the model, if we manage to reach an accuracy around 90% just by rerunning the model training several times in a row, given that it will keep the updates and start the new training from the previous accuracies, does this amount to training the model over more epochs? Is this a good practice, will training the model in this way help it to generalize?

[@parthsarthijoshi6301]

Sir I am 14 year old studying in class 8 . My dad is 55 year old has recently started learning data science . He is in government job .I was motivated to learn data science and has completed Data Analyst course. In the past when I was in class 4 I completed C++ programming language. I am a regular listener of your videos on data science. I want to face your mock interview exercise . Kindly respond

[@porikanithyanand7924]

Eri pooka...

[@sarderiftekharahmed5558]

Can I use MUSIC data set here? like FMA

[@knowledgedoctor3849]

Since Class 9 is not dog barking?

[@krishnaik06]

yes it is a wrong classification

[@knowledgedoctor3849]

@@krishnaik06 is it possible to use RNN OR LSTM to make it better?

[@deepali-e6f]

Can we use similar preprocessing steps if I have audio of songs.. I am working on Music genre classification, so we can use the MFCC concept ?

[@sand9282]

Sir isn't this problem a multi-label problem rather than a multi-class classification problem? I am asking because I can see 10 features in your target variable y. So in the case of multilabel problem won't you be using a 'Sigmoid' function rather than the 'Softmax'?

[@faresbecheikh7052]

Please how to plott the Confusion Matrix of this Project ?

[@muhammadsaad872]

Hey. Amazing video. I have one query though you used Relu as the activation function in the first layer, whereas the data is scaled between -1 to 1 so wouldnt that be a problem since the first layer will only take values from zero onwards?

[@raj-nq8ke]

No, ReLU is an activation function. It will not affect the normalized data. Data will go through a CNN filter and then the activation function will take action.

[@santhosh20071993]

Yes, You are right. It has to be TanH. People often applies activation function without understanding the math. Also since this is a classification problem, TanH activation is ideal. Because softmax/sigmoid is mathematically same as TanH equation

[@areezabhanbhro6005]

What type of ANN is used here? Either DNN, CNN,RNN,MNN ? Which one is he using here

[@AbhinavSingh-ud8dc]

where is this audio_classification.hdf5 file present?

[@louerleseigneur4532]

Thanks Krish

[@abeljacob611]

AttributeError: 'Sequential' object has no attribute 'predict_classes' model.predict_classes(prediction_feature) This code is no longr supported I am getting this error and when i use the alternate code for this my predictions go wrong anty tips on what i can do please?

[@yogeshwarshendye4857]

I have a question. You just trained the model two times each time with 100 epochs. Does that over-fit your model or what?

[@marianagesheva9560]

Great series. One thing - I am getting an error at the end. AttributeError: 'Sequential' object has no attribute 'predict_classes' . I cannot find out the predict_classes

[@bihahazlan3393]

me too , how to slove it?

[@2dFunToonAnimation]

@@bihahazlan3393 I’m also getting the same

[@apoorvadesai604]

Hello sir, how to add confusion matrix in audio classification.

[@liorhason2621]

Hello Krish, Thank you for all of your useful videos. I am working on a project with 4 classes in .wav form. How do I convert the information to a csv file?

[@HealthyConnection64]

Hello, this is a great video! Is it possible to test this using other wave files that are not in the dataset. And could this detect certain sounds in a longer wav file

[@samiulbasirrajib3038]

Hello, I got only 10% accuracy. What is the problem of my code?

[@kashyapibisen1769]

hey did you get any solution i am having same issue

[@ravitanwar9537]

dropout to reduce effect of overfitting

[@ravitanwar9537]

@@jonsnow8297 ?

[@16_surajkumar79]

how to label peak of wave

[@kashyapibisen1769]

hey there! i am getting same accuracy for every number of epochs i.e. 0.10475099831819534 could you please tell me what should i do in this situation.

[@mufseeramusthafa2170]

I got only 25 % accuracy

[@sagarkhule6439]

Try again using CNN

[@2dFunToonAnimation]

Increase no. Of epochs

[@anujgupta9081]

I'm getting only 10% accuracy.

[@sagarkhule6439]

Try using CNN

[@2dFunToonAnimation]

Increase epochs