Before starting PixVerse, I prototyped a NLP lyric writing helper with a markov-chain language model. Sadly, it didn’t seem to offer much practical help when I actually tried to write some lyrics with it. Back on the shelf it went, until recently.
I came back to song writing, but this time on iPhone. I changed the goal to be: “A rhyming dictionary that works really well for songwriters.”
I looked at the existing rhyming dictionaries on the web and iPhone. A lot of them really don’t work well – you get all these words that don’t really fit thematically with what you’re trying to write. No soul, just obscure terms that happen to rhyme.
So it was time for a new approach: Use a lot of interesting data and simple algorithms.
iRhyme constructs its rhymes database from a simple natural language analysis of a very large song archive (corpus). Think of iRhyme as the collective knowledge of songs and songwriters condensed into a 16 MB iPhone app. Most of the effort went into packing all those rhymes in a easily retrievable format.
It was very cool to see the results of the analysis. With some tweaks, iRhyme had linked over 1.5 million relevant words and rhymes together, with some pretty canny free-associative rhymes emerging from the data. iRhyme has over 50,000 unique words, all used in real songs.
For now, you can buy iRhyme for $0.99 on iPhone. It’s a pretty good value compared to the cost of an “old world” printed 100,000 entry rhyming dictionary. iRhyme also fits better in your pocket (the entire database is stored locally).
I am also currently working on upgrading iRhyme’s database, increasing iRhyme’s depth even further, which will part of an upgrade once I can figure out how to pack all the new data into the 10MB over-the-air App Store limit.