Barefoot Belle And The Pursuit Of Music

Barefoot Belle is now my favorite song ever, the vocals, and atmosphere, the raccoon, the crown, and living up in the trees, all just perfect.

While I have not been conducting a real search, I have spent a few days building up my song generator.

My biggest finding is that you can’t just specify bluegrass or country, AI accepts three important numeric values cfg, topk and temperature.

When I leave them at factory settings, I get bluegrass or country, a very normal version of it.

Very soon, you will hear the same exact normality everywhere, or at least everywhere that does not want to pay royalties.

But when you ask your favorite AI how to adjust values for a particular model, it will do a bit of research and tell you what gently adjusting those numbers will do.

I am convinced that I can create songs as beautiful as Barefoot Belle, but based on about 30 tests I conducted today.

I will never be able to reproduce Barefoot Belle exactly, AI and I agree that the song is a little miracle.

And perhaps more importantly exactly what I needed, to take a closer look at fine tuning pretty song generation.

It is very interesting that I can create pop and country songs, that fell the same, that seem to share a soul.

And I am glad that I can change some settings, and gravitate closer to lo-fi older county music as it is very pretty.

It is of some note, that when it came to examining the song generator in more depth.

I switched from graphic user interfaces, to text based command line, where I could adjust settings with grater ease without graphics getting in my way.

Some of my earlier experiments had a primitive dollar store instrument sounds, that is because my songs are lyrics heavy have no room to develop melodies.

And I did nothing to enhance the instruments, or give the AI reference sound, now I have a neat AI generated reference, and a studio sound enhancer.

I also added a song splitter that divides vocals from drums and melody, and here I can test lyric pronunciation.

And this will be my closing thought, my most important finding, do not use local AI for lyric creation, use the most fanciful free AI you can find.

A couple of songs will not overwhelm its free memory, at all, and the lyrics make all the difference in the world.

Keep your song in text, but ask your AI to translate at least, the commonly mispronounced words into a phonetically spelled version.