Music Modeling and Generation
Music is the ultimate language. Many amazing composers throughout history have composed pieces that were both creative and deliberate. Composers such as Bach were well known for being very precise in crafting pieces with a great deal of underlying musical structure. Is it possible then for a computer to also learn to create such musical structure?
Automatic Music generation is one of the hot topics in AI Research nowadays, with many big companies like Sony, investing in reviving old classics from Beatles, Michael Jackson etc. The problem of music generation is similar to that of Language/Text generation, but is much more difficult. It is difficult to generate likable/good sounding music. Generating music with long-term structure is one of the main challenges in the field of automatic composition.
Over the years, many different techniques have been proposed, some relying on standard NLP techniques, like N-Grams, Hidden Markov Models and others on more complex Deep-Learning based methods. Standard NLP techniques, like constructing language models from musical notes, is often not able to capture the essence of music, wherein musical notes, far-apart from each other may be related and complementing to each other.
A popular approach to solve this problem, is the use of a special kind of Recurrent Neural Network called LSTMs. Long Short Term Memory networks – usually just called “LSTMs”, are a special kind of RNN, capable of learning long-term dependencies. They were introduced by Hochreiter & Schmidhuber (1997). They work tremendously well on a large variety of problems, and are now widely used for research in Music Generation.
LSTMs are explicitly designed to avoid the long-term dependency problem. Remembering information for long periods of time is practically their default behavior. All recurrent neural networks have the form of a chain of repeating modules of neural network. 
Different ways for encoding music have been proposed, for use in LSTM; Some works have proposed using instrument notes, for constructing language vocabulary and subsequent language. Some works, on the other hand have focused more on converting raw-music files into text representable form for further processing.
Music generation is still an active area of research. The following are some areas, where more work is required.
- Creating music with musical rhythm, more complex structure, and utilizing all types of notes/Larger vocabulary
- Creating a model capable of learning long-term structure and possessing the ability to build off a melody and return to it throughout the piece
Some popular systems for Automatic Music Generation are MorpheuS, GRUV etc.
Comments
Post a Comment