Audio Classification with Machine Learning

Explore audio classification using machine learning in this 43-minute EuroPython Conference talk by Jon Nordby. Dive into the world of sound analysis, learning how to convert audio into spectrograms and apply convolutional neural networks for classification tasks. Discover practical applications in speech recognition, music analysis, medical diagnostics, manufacturing quality control, and animal behavior studies. Gain hands-on experience with Keras and TensorFlow frameworks, and learn valuable techniques for achieving usable results with limited data, including transfer learning, audio embeddings, and data augmentation. Covering topics from digital sound representation to environmental sound classification, this talk provides a comprehensive overview of audio classification techniques, suitable for those with a basic understanding of machine learning and digital sound.

Intro
THIS TALK
APPLICATIONS
DIGITAL SOUND PRIMER
AUDIO MIXTURES
AUDIO ACQUISITION
DIGITAL SOUND REPRESENTATION
SPECTROGRAM
PRACTICAL EXAMPLE
URBANSOUNDSK
MEL-FILTERS
NORMALIZATION
CONVOLUTIONAL NEURAL NETWORK
AGGREGATING ANALYSIS WINDOWS
DEMO
ENVIRONMENTAL SOUND CLASSIFICATION ON
TIPS AND TRICKS
DATA AUGMENTATION
TRANSFER LEARNING FROM IMAGES
AUDIO EMBEDDINGS
ANNOTATING AUDIO
SUMMARY
MORE LEARNING
QUESTIONS
TAGGING