AudioSet by Google VS Photosounder

The AudioSet database (https://research.google.com/audioset/) is a large collection of audio clips with corresponding spectrograms and annotations for use in machine learning research. Source: over 1 year ago

Hello everyone - I'm a composer currently working on a project to make use of Google's enormous dataset of annotated 10 second audio events taken from Youtube videos, Audioset. Source: over 1 year ago

Photosounder exists. https://photosounder.com/ The problem is, if you listen to the examples, they sound extremely robotic. Source: 12 months ago

Now that I had the opportunity to listen more closely - the second one reminds me of spectral editor tools such as https://photosounder.com/ . Specifically https://www.youtube.com/watch?v=W8MCAXhEsy4 - it's all about picking the right image. I think you can also do this with Harmor - load an image and resynthesize it. Since that gives you a lot of sinewaves, I'd say your theory is correct :). Source: about 1 year ago

Also worth noting that https://photosounder.com/ as a spectrogram editor has been around awhile. Source: over 1 year ago

If you read an image file, then each pixel can be represented as a sample value, and then you're reading it line by line. However, another approach is additive synthesis - where each line of an image is turned into a harmonic, and the brightness of each pixel is its amplitude. Source: over 1 year ago

Start with subtractive, then work your way up to basic FM. For additive synthesis, people use image editors. Source: over 1 year ago