Esperanto AI

What better thing to do with old lyrics than to do un-singable versions of them in an obscure constructed language as blues/country songs? 

As fun as it was, it's still not real. There's no singer who's going to perform in Esperanto. No one speaks in Esperanto, let alone easily sings it.

Other thoughts in retrospect: 

• I find the translations to be clumsy and excessively wordy.  "Kapitulacu al la forto". Try spitting that out while playing guitar, but AI can do it easily.
• The sound of the words is interesting--at once Spanish, Italian, and Greek. If the language had a place, it would be the lands around the Mediterranean.
• Now that we can easily "re-write" songs with AI, it's interesting to experiment with different genres.  I use the term "write" loosely because there is no writing involved. It is copying and pasting, and writing an LLM prompt. But in the process of finding music that sounds good, we trivialize the lyric.
• If you were to do this manually, say, 30 years ago when these lyrics were written, you'd have to find someone to translate them, then write melodies. Since I don't speak the language, it would be impossible to sit down with an instrument and sing against it. But AI can do that, but it's a soulless mechanical process--something I sometimes find hollow and creepy. The music itself works as a kind of addictive confection which hides the fact that the sugar isn't good for you.
• As slick as the productions are (the guitars sound nice), you could take any old text like lorem ipsum and make it an Esperanto song, which obviously is a "profane" activity. I'm using my lyrics, which I sang at one point, so it is from me, but the vocalists don't understand the lyrics. What it understands is the patterns in language that get mapped over the patterns in the individual stems. The interpretation arises from tokenized text, not a human reflecting on their meaning.
• There was an interesting study that was done a few years ago, "Song lyrics have become simpler and more repetitive over the last five decades", on the complexity of lyrics. There's a difference between those who write scientific papers about lyrics and those who actually write and sing them, which has an interesting connection with AI-generation of songs where the "singers" are just LLMs shuffling patterns to match riffs. Personally, when I work on a lyric, I'm focusing on how the quality of words to be used musically, especially at the rhythmic level. How syllables are mapped over note values is crucial. I think lyrics should be as economical as possible, which is the issue with translations from English because you're "deracinating" the natural speech rhythms. With the Esperanto lyrics, the music sounds fine on its own, with the vocals just being another instrument. If you don't understand the words, then it's essentially an instrumental. Also, I think lyrics can be cryptic. They don't necessarily have to mean something--precisely because of their percussive role. I had heard somewhere that Peter Gabriel still doesn't know what Lamb Lies Down On Broadway means. I love that things can remain ambiguous. 
• The music is very often inappropriate to the tenor of the lyric, even though it sounds otherwise. My version of Volo de la Mondo (Want of the World) was inspired by Oasis at the time and had a cut-time feel with atmospheric guitars, not a bouncy groove. The AI version is still cut-time, so it did detect that feel in the words, which is interesting. The groove is impeccable and would take a good rhythm section to reproduce that kind of "chugging" rhythm, not to mention the ability of the singer to negotiate the words.
• Esperanto is a very odd language, even though it has its moments. It must be very hard to sing because it is very consonant-heavy, but we'll never know because barely anyone speaks it. (I recall Kurt Elling doing a few tunes in Esperanto).
• One of the good things about music LLMs is that they can suss out any phonemes, so it's good for reviving lost languages--through song. 
• Some of these songs already have other AI versions, like Force of Will, Blood River, The Inner King, Don't Lose Your Heart, and Never the Same River Twice, which never had a version played by me. It's always been "natively-AI".

Explore deeper with the LLM