TensorHero ML

We are creating a system capable of ingesting raw audio and outputting playable Guitar Hero charts which match the musical structure of that audio. Drawing on work from music information retrieval, machine learning, and signals processing, we are using a PyTorch Transformer model and various data manipulation techniques to create a chart for any audio file that plays just like any human-created GuitarHero chart. 

Our team explored many preprocessing steps and metric evaluation techniques to improve the model outputs, and generate charts that provide a better playing experience. The model was operationalized in partnership with the Front-End team, who created a website that provdies an interface for uploading a sond and receiving back the playable chart file.