Music is a temporal art, which unfolds over time. Unlike painting, which is a spatial art, can be seen entirely immediately. Because of its temporal nature, music can tell more things than other arts do. Composer can express his feeling and tell his story through music. Song is a kind of music that contains lyrics. Song can express more things than a plain music does. People want to have their own songs for some special events such as congregation and wedding. Composing a plain music is hard. Composing a song is even harder because of the presence of the lyrics. People may seek help from computer to compose their own songs. The study of using computer to compose music is called Algorithmic composition.

There is an algorithmic melody composer called “T-Music”, which mines the co...[ Read more ]

Music is a temporal art, which unfolds over time. Unlike painting, which is a spatial art, can be seen entirely immediately. Because of its temporal nature, music can tell more things than other arts do. Composer can express his feeling and tell his story through music. Song is a kind of music that contains lyrics. Song can express more things than a plain music does. People want to have their own songs for some special events such as congregation and wedding. Composing a plain music is hard. Composing a song is even harder because of the presence of the lyrics. People may seek help from computer to compose their own songs. The study of using computer to compose music is called Algorithmic composition.

There is an algorithmic melody composer called “T-Music”, which mines the correlations between the melody and the lyrics in song database and uses these correlations to generate a melody for the user-input lyrics. The correlations are represented in frequent patterns (fps).

This thesis studies the ways to enhance T-Music. We have defined a new way of how a pattern is considered to be frequent. By using the new definition, the fps can represent the correlations more realistic.

The original mining algorithm can only mine the fps from songs. The lyrics must be presence. Two new algorithms are proposed to mine the fps from plain music.

In other to use the fps mined from songs with lyrics in the language L₁ to compose a melody with the lyrics in another language L₂, a tone mapping is required from L₂ to L₁. An optimal mapping algorithm has been proposed so that T-Music can use the fps in L₁ the best to compose a melody for the user-input lyrics in L₂.

In other to impose genre on the generated melody, the characteristic of different genres have been studied and are included in T-Music. T-Music can generate a melody for the user-input lyrics in some particular music genres such as traditional Chinese music and dance music.

A business plan based on T-Music has been included so that the uniqueness and values of T-Music can be well presented.

In addition, a comprehensive introduction of the background knowledge of music and language are included so that readers with different backgrounds can understand this thesis well.