Developed an AI model as part of an independent research course that understands the positive/negative sentiment in a given piece of text and generates music that fits the sentiment!
The AI uses a Sentiment Analysis model to first understand the sentiment of the text and then based on the sentiment, it picks a pre-trained music generation model to generate music specific to that sentiment.
The Sentiment Analysis model consists of a bi-directional GRU Recurrent Neural Network (RNN) that creates an embedding of the given text and then passes this embedding into a fully-connected Neural Net that classifies the sentiment.
The music generation model generates uses a GRU network to generate the ABC-notation which is then converted into a .midi file using the abc2midi library for playing.
More information can be found at our GitHub repository.
Sentimuse used to be hosted on a website, but it is no long active (Azure free trial expired 😔).
Some sample music the AI had generated:
Positive sentiment music:
Negative sentiment music: