matt-heff / ai-beat-maker

Generate beats out of given samples

AI-Beat-Maker

Description:
Creating music has become easier with remix generators and separate drum line creators, but there is still no solution for automatically creating the desired combination. We use a combination of machine learning and DSP methods in order to automatically generate a remix out of a given sample, and sync the result with a generated drumbeat created by using a trained neural network. Thus, we have created a beat-making program to give our users the ability to create a brand-new Hip Hop beat in the simplest touch of a button, without the need for advanced and complicated music studio applications.

Running Instructions

1. Clone this repo
2. Activate virtual environment
3. Install requirements

pip install -r requirements.txt

4. Add the sample you wish to remix into the 'samples' directory
5. Run main.py
   Parameters required:
   --track -> samples title with no extension (must be a wav file)
   Optional parameters:
   --drums -> True if you wish to generate drums for the remix else False (default=True)
   --bass -> True if you wish to generate bass for the remix else False (default=False)
   --k -> Number of clusters you wish to split your sample (default=2)

For example:

python main.py --track=track_name --drums=True -- bass=False --k=4

If you would like to retrain the LSTM Drum-Generator, run:

python drums_generator/lstm.py

You can then find the new model weights in the 'drums_generator/weights' directory.
All that's left is to update the predict.py file to load your new weights.