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

Course Chatbot

• We have a course chatbot, the Deep Learning for Audio Tutor

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.

Online Speech Translation

The html file speechtranslator_grok.html is a browser app, which records voices from the microphone, recognizes them, and translates their language, for instance from German to English, continuously, and prints it out as text on the screen, and which then can be optionally saved to a file. You can open it using github.io with this link:

speechtranslator_grok.html

but it looks better when the html file is downloaded and then opened with the browser. It uses webkitSpeechRecognition.