Mass Audio (tm) - Machine Learning with L3 - Look Listen Learn More
Mass Audio (tm) is a module in the school of Techbotism Audio production
This is a work in progress as I figure out Machine Learning from scratch. I literally need to google at least one word per sentence in everything I read.
Over the years we've collected over 8,000 albums. Since we joined Bandcamp 18 months ago we've amassed another 10,000 albums.
928 gigabytes of samples,
376 gigabytes of multisamples,
84 gigabytes of Sampl CDs,
257 gigabytes of djs mixes ,
164 gigabytes of Radio Stuff
I want a machine Learning AI thing to analize all this audio break it down into units that are categorisable. Then when I choose a piece of audio, it recreates new pieces of audio isungmy chosen one as a staring point and replacing each of the untis within the audio with a similar unit based on whatever options are availble. This is then broadcast as the sum of all audio eminating from the planet and bounced back via the hijacked EMC23 pirate satellite
So what are we talking here? Well Audio Word Embeddings (AWE) are patterns for words spoken at variable speed, so we're talking dynamic time warpng. (two people start walking beside each other at slightly different speeds- the defining pattern remains the same). So with L3 we are talking Deep Audio Embeddings or patters for sounds. So while AWE might be good for voice recogition DAE might be good for sound recognition, pitch recognition and maybe instrument separation in a mix I dunno. this is my first day at school.
What I woud hope to do is break my Mass Audio (tm) collection into DA Embeddings and from them analyse and auto generate ambient, experiment and avant garde music. I would hope to go deep into convolution techniques
Step 1: Install Python 3.7 ( 3.9 prevents h5py-3.2.1.dll )
Step 2: Install tensor flow 1.14 (not tensorflow 2)
pip install "tensorflow<1.14"
Step 3: Pip install openl3
pip install openl3
Open L3 Tutorial: