search

cognitive-catalyst / watson-beat

Other
102 stars 46 forks source link

Creating actual track #27

Closed ghost closed 4 years ago

ghost commented 4 years ago

Hello! Thank you very much for making this library available. I am not a musician, but I have a basic understanding of music, so perhaps that is why I am not sure how to go about creating a track using this library. I am interested in creating a music that I could eventually use to go along with some lyrics. So, I am wondering if we can we use any midi file that is more than a minute long? Specifically, could an input midi file that long help create midi file(s) so as to create a music that could be used for a song? I did run an example that you have under Ready to Run, which is good to understand the elements of the code, but I am looking for some guidance on how I can create my own tracks like those here (by Richard Daskas). Although I did go through the Video Tutorials that helped understand the post-processing with the output midi files, I would appreciate your kind help if you could describe the process on how to go about creating actual tracks (like this) using this library that are few minutes long.

amchaney commented 4 years ago

Hi, You need to read the README.md file https://github.com/cognitive-catalyst/watson-beat

The command python wbDev.py -i Ini/ReggaePop.ini -m Midi/mary.mid -o /Users/Moe/midifiles/ (for example) is when the midis are actually created. To adjust the length/number of sections/tempo/etc edit the ini file. To get started, just use one of the midi files checked into the repository.

As far as creating a "song", there is an example I did with Tayrn Southern https://www.youtube.com/watch?v=bLHid5yyZkM using this code, but there was a TON of post production involved. The code did create the stems, and Tayrn wrote the lyrics.

ghost commented 4 years ago

Hello,

Thanks for your comment, @amchaney. I did try some examples from the repository. Can we use any midi file that is more than a minute long? Would it be computationally advisable in a sense that can the code easily handle that much training data? Thank you!

amchaney commented 4 years ago

I have tried different lengths and complexities of midis, I find the short and simple ones work best with this algorithm. The best thing to do is just run the code and try it and see what you get. I don’t think the code imposed a limit on the midi, but the idea is it is a melody hook and the algorithm riffs on that.

ghost commented 4 years ago

Thanks @amchaney!