단풍 <danpung>

단풍 (dahn-poong)

Danpung is a rhythm game and autumn walking simulator, based on how I enjoy my music while walking around UNC's campus.

HOW TO PLAY: Just download the game, and by hitting "Play" you can play the game based off single track of music this game has. Hit the up and down arrow keys to hit each incoming beat, and you can see your score at the end of the level.

CUSTOM MUSIC: The example song in the game is personally annotated by me, but the big part of this game is that you can upload your own music!!! (seriously I spent like 7 days of the game jam getting this ML model set up to recognize beats and its a little funky but I'm proud of what I learned!) ANYWAYS, if you hit the "Open Music Folder" button, you'll be directed to a folder on your computer where you can put your own mp3 files, and in-game just click the name of the current song until you're at the one you want to play. After that just hit "Generate Song", and the game will freeze for about 2-3 minutes depending on the song length as its machine learning model generates a music track. After that just hit Play and you're good to go!

TECH RUNDOWN: For more on the custom music and the model that does it, inside the game I added functionality to annotate your own music files to create a playable track from it, and put a script in the game to train two ML models based on the portions of the music I annotated. Both those features are locked behind a "dev mode" toggle or are commented out in the release, and you can see both of those features in the BeatMapper.cs and MLHandler.cs files in the github repo. The BeatMapper class handles the saving and loading of the song tracks on the computer, as well as turning those .beattrack files into .tsv files to be used by the ML model. The MLHandler class trains, saves, and uses two binary classification models to determine whether or not any point in the mp3 file is a beat, and whether it's an upper/lower beat. I trained the models mostly on how I annotated the default song, and the models currently act a little crazy with end-of-song silence, but by annotating more genres of music this model can still be improved.

CREDITS AND INSPIRATION: Okay so basically everything you see in this game was made by me (save for the font) since Thursday. However, HUGE PROPS TO MY FRIEND GEETHAN for making the track pre-installed in the game! Could not have made that amazing track myself, many thanks to him. Other than him, I'll list some different tutorials that helped me set up the game, create the (blender & machine learning) models, and get this thing working. Also random but the inspiration I got for this project I had while driving back from Virginia Tech through the mountains and there was this one really really good autumn foresty mountain skyline view that I stopped that gave me the inspo for this.

NAudio Library (almost made this unity project not export tho)

ML.net Beginner Tutorial (didn't exactly use this but this was definitely my baseline)

Simple Blender Character Tutorial (big big big help for my blender character & skills)

Simply Rounded Font (the font you see like three times)

I also used a few songs to train the ML model in the game, those being Geethan's song, MASCARA by King Gnu, Kick, Push by Lupe Fiasco, Reckless Driving by Lizzy Alpine and and Love by Keyshia Cole.