Skip to main content

· 38 min read

What follows is an exact repost/mirror of Steve Yegge's 2012 post "Notes from the Mystery Machine Bus" from Google+ on the software risk aversion axis. It's not even accessible from the WaybackMachine. Enjoy.

· 4 min read
Luis Gomez

The Wrong Way​

There are SEVERAL guides online telling you how to normalize audio to provide a great listening experience by using FFmpeg's "loudnorm" filter. Most walk you through a simple process called "track normalization." You set a goal loudness, and adjust every piece of audio up or down to match that loudness. This is evil and immoral and I'll tell you why:

Loudnorm changes the way things sound!

Loudnorm crushes dynamics!

· 4 min read
Luis Gomez

Let me get this off my chest right off the bat: Keras is all you need for 90% of neural network applications. Before jumping in, you should understand what is going on with neural networks(watch 3Blue1Brown's videos).

Ideally, you can understand all the math behind them and code a basic one from scratch (read Nielsen's Neural Networks and Deep Learning).

But for most cases, even understanding is unnecessary with all the pretrained models frameworks ship with nowadays.

But, assuming you're not a n00b and you need to do more beyond just classifying 2D images or recognizing characters, here are the Deep Learning/ Neural Network frameworks can best fit your use case.

The Only 4 Deep Learning Frameworks That Matter​

· 3 min read
Luis Gomez

Why?​

I wanted a faster, more efficient way to make music.

My solution? Stick a Launchpad in a guitar

Sadly, no electric guitars I found had the room for a launchpad.

My solution (pt 2)? Build my own guitar with correct dimensions.

My advice? DON'T DO IT. It was really hard and time consuming.

Unless you NEED to do this to get specific dimensions & realize some crazy idea of your own, just buy a nice used guitar. You'll still get a lot of instrument for your money, and SAVE TIME.