Audio ML fundamentals, packaged as notebooks, slides, and YouTube links

What it does

This repository is the companion material for a YouTube course on audio signal processing for machine learning. It bundles Jupyter notebooks, PDF slides, and sample .wav files into a structured curriculum that starts with sound wave physics and ends with extracting Mel spectrograms and MFCCs in Python. Think of it as a semester’s worth of lecture notes, except the professor is a playlist and the homework runs in a browser.

The interesting bit

Most tutorials hand you a librosa one-liner and call it a day; this one stops to explain why the Fourier transform uses complex numbers and what a band energy ratio actually is. The maintainer also did the rare courtesy of modernizing the codebase to Python 3.11 and current dependencies in a v2 release, while archiving the original course version in a legacy branch for anyone following along pixel-for-pixel.

Key highlights

• 23 chapters covering foundational acoustics, time-domain features, DFT/STFT theory, and frequency-domain implementation.
• Every chapter links to a YouTube video; most include both PDF slides and a runnable Jupyter notebook.
• Updated to modern Python, librosa, NumPy, and Matplotlib, with a preserved legacy branch for video parity.
• Includes an audio_resources/ directory with .wav files used in the examples.
• MIT licensed.

Caveats

• The v2 code has been updated for “modern best practices,” which means it may not match the original videos exactly; purists will need the legacy branch.
• This is a learning guide, not a reusable library or framework, so expect educational code rather than production abstractions.

Verdict

Grab this if you are an ML practitioner who has been treating audio features as black boxes and wants to know how MFCCs are actually computed. Skip it if you are looking for a drop-in audio preprocessing pipeline or a model zoo.