Machine Learning for Audio Signals in Python

Prof. Dr. -Ing. Gerald Schuller
Jupyter Notebooks and Videos: Renato Profeta

Applied Media Systems Group
Technische Universität Ilmenau

Content

01 Neural Networks Basics - Detector:

- Introduction

- Neural Networks as Detectors

- Fully Connected Layer

- Activation Functions

- Optimizers

- Python PyTorch Examples

02 Neural Network as Function Approximator, Regression:

- Introduction

- Function Approximation

- PyTorch Example: Shallow Network

- Deep Function Approximator

- PyTorch Example: Deep Network

03 Neural Networks for Classification:

- Introduction

- MNIST Dataset

- PyTorch Model

- Cross Entropy Loss

- PyTorch Example

- Unknown Test Image

04 Neural Network Detector for MNIST Digit Recognition:

- Introduction

- One-Hot Encoding

- PyTorch Example

05 Convolutional Neural Networks:

- Introduction

- A 1-D Signal Detector

- An Audio Predictor

06 Convolutional Autoencoder:

- Introduction

- PyTorch Audio Convolutional Autoencoder

- Effects of Signal Shifts

07 Denoising Autoencoder:

- Introduction

- Experiment 1 with stride=512

- Experiment 2 with stride=32

08 Variational Autoencoder (VAE):

- Introduction

- Posterior and Prior Distribution

- Kullback–Leibler Divergence

- Variational Loss

- Lagrange Multiplier

- Variational Autoencoder Experiments

09 Recurrent Neural Network (RNN):

- Introduction

- Infinite Impulse Response (IIR) Filter Structure

- IIR Python Implementation

- IIR Implementation using RNN in PyTorch

- Training the RNN

YouTube Playlist

Requirements

Please check the following files at the 'binder' folder:

environment.yml
postBuild

Note

Examples requiring a microphone will not work on remote environments such as Binder and Google Colab.