13- Implementing a neural network for music genre classification

Показать описание

In this video, l implement a music genre classifier using Tensorflow. The classifier is trained on MFCC features extracted from the music Marsyas dataset. While building the network, I also introduce a few fundamental deep learning concepts such as binary/multicalss classification, rectified linear units, batching, and overfitting.

Video slides:

Code:

Interested in hiring me as a consultant/freelancer?

Join The Sound Of AI Slack community:

Follow Valerio on Facebook:

Valerio's Linkedin:

Valerio's Twitter:

Рекомендации по теме

Комментарии

7:19 That's brilliant. It's one thing to be good at memorizing and API but you're also a genius! This is what makes a good programmer. Super excited to learn more from you, this is what should be in YouTube trending!

geofox

Very nice illustration Valerio, specifically at the end where you showed the overfitting.

mostafahasanian

Perfect recap after DLS Course by Andrew Ng. Your videos are so awesome as Coursera's. Thank you!

chipotle

Great elegant coding and clear instruction! One potential issue here is the data leakage problem. Since each mfcc array is generated from 1 out of 5 of total segments you split a tack into. This means that naive usage of train_test_split will result in a situation that test set and train set share different segments of the same track and test accuracy will be overestimated during model development phase. Better split of data should be done on a track level but not segment level.

spkt

Super amazing, full pack playlist for acoustic deep learning! Thanks alot Valerio for this!

raghavrawat

Thanks a Ton! Your videos are really instructive Dr. Velardo. Looking forward to more videos/Lectures from you.

jainrohit

You are love man. Can't w8 for more video on deep learning music

smilebig

i watched both video 12 and 13 and they worked. So my question is how can i see classification of songs? i could not see here, which of your videos shows the result of classification

berkinoztekin

Hey dude,
congratulations for your class/share link.
We keeping walk. Chers! :)

i_am-ki_m

What is the error formula when you input the batch or all inputs? Thank you.

nhactrutinh

I replaced sigmoid with ReLU in the simple MLP network (covered in video # 9) & the predictions started coming to be 0 in many cases. Not sure what is causing this - but do we see this being an issue with audio data - MFCCs can have -ve values and that may make h -ve.

maulikdave

Awesome contents. Thank you and btw did your machine have GPU when training in this video?

ngocminhphung

Amazing, it trains just 127 samples on my mac, but trains the complete samples on a windows machine. Help would be highly appreciated

shams_ad

Hi...if we are loading one data point and then performing forward pass and back pass how can this be faster? Essentially we are sequentially reading data in RAM . Loading data in RAM in larger batch should be faster. Am I missing something..

physicsmadness

I suppose I can't load the whole json file.

When I print(inputs.shape) The output is (200, 130, 13) Because there is 2 files in each folder of my reduced dataset.

When I tried the same fot the whole dataset, with 100 tracks in each folder, The output is (10000, ) I was expecting to see it be something like (10000, 130, 13) .

So, am I right? Is my data not complete? Because the size of my data.json file is 647MB? If this is the case? How did Valerio's code read such a big file without error?

Thanks for any help in advance.

SabriCanOkyay

Why when we, theoretically, use a full batch, we need just one epoch?

amitbenhur

Hello. I am using spectral centroid for audio classification. But I have one problem. When I use MLP for classification my val accuracy is constant from epoch 1. What is going on?

bujipaji

i have question, i previous you said we will have 5 segments like we are dividing so 30 sec video divided into 5 segments = 5 segments with 6 secs of data and here you are saying u made 10 segments

muntazirmehdi

Hello Valerio, thank you so much for your videos. I can't load all the 6997 samples when I trained, I only have 219 samples. How could I solve it??

hoangphuc

sir how can do speaker change detection for diarization

Liya

13- Implementing a neural network for music genre classification

13- Implementing a neural network for music genre classification

13-Neural Network Implementation From Scratch in Python | Machine Learning | Deep Learning

But what is a neural network? | Deep learning chapter 1

Implement Neural Network In Python | Deep Learning Tutorial 13 (Tensorflow2.0, Keras & Python)

Create a Basic Neural Network Model - Deep Learning with PyTorch 5

PyTorch Tutorial 13 - Feed-Forward Neural Network

Here Is How Neural Network Work... | #neuralnetworks #chatgpt #usa #newyork #physics #demo #science

Neural Network from Scratch | Mathematics & Python Code

🤖Will Artificial Intelligence Make You a Millionaire from Trading?! (Shocking Truth!)💰📈

Learning an image using neural network #ai #machinelearning #deeplearning

Programming for AI (AI504, Fall 2020), Practice 13: Graph Neural Networks

Transformer Neural Networks Derived from Scratch

Neural Network From Scratch In Python

Implementing Neural Networks in Python

Gradient descent, how neural networks learn | Deep Learning Chapter 2

Gradient Checking (C2W1L13)

Day 13 - Design your First Neural Network

I Built a Neural Network in C++ from Scratch (NO Tensorflow/Pytorch)

Convolutional Neural Networks with TensorFlow - Deep Learning with Neural Networks 13

Creating Photographs Using Deep Learning | Two Minute Papers #13

2. Implementation of AND function using PERCEPTRON model | Artificial Neural Networks

Neural Network learns Sine Function with custom backpropagation in Julia

Lecture 13 (EECS4404E) - Neural Networks (Part IV) - Implementing Neural Nets

Wear volume prediction of AISI H13 die steel using response surface methodology and a... | RTCL.TV