Automatic generation of music in various styles   日本語 (Japanese)

Music generation based on style conversion

There are various styles in music such as classical music, jazz, and J-pop. Music style conversion is a technique to generate a musical piece in a target style by using the syntax structure of a reference piece. In this way, we can automatically generate a variety of musical pieces in various styles, with different combinations of a reference piece and a target music style. Moreover, it enables us to generate high-quality musical pieces that hold structures found in human-created music such as repetitions and contrast.

Style variations on Ode to Joy

Examples of style conversions based on Beethoven's Ode to Joy


Style conversion examples using musical score data

The following examples are automatically generated using manual transcriptions of reference pieces.

Dragon Night by SEKAI NO OWARI

Melody transcription of the reference
Piece in J-pop style
Piece in J-pop version (no. 2)
Piece in pentatonic scale style
Piece in enka style
Piece in jazz style
Piece in classical music (minor mode) style

TSUNAMI by Southern All Stars

Melody transcription of the reference
Piece in pentatonic scale style
Piece in Okinawa music style
Piece in enka style

Family Party by Kyary Pamyu Pamyu

Melody transcription of the reference
Piece in jazz style
Piece in enka style
Piece in J-pop style

C.O.S.M.O.S. by Sandaime J Soul Brothers

Melody transcription of the reference
Piece in pentatonic scale style
Piece in enka style
Piece in J-pop style
Piece in J-pop style (no. 2)

More examples coming soon ....

Style conversion examples using automatic transcriptions of audio data

Coming soon ....

References

Please see the following papers on the principle of automatic style conversion and technical details.

  1. Eita Nakamura, Kentaro Shibata, Ryo Nishikimi, Kazuyoshi Yoshii
    Unsupervised Melody Style Conversion
    Proc. 44th IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 196-200, May 2019.
  2. Ryo Nishikimi, Eita Nakamura, Masataka Goto, Katsutoshi Itoyama, Kazuyoshi Yoshii
    Scale- and Rhythm-Aware Musical Note Estimation for Vocal F0 Trajectories Based on a Semi-Tatum-Synchronous Hierarchical Hidden Semi-Markov Model
    Proc. 18th International Society for Music Information Retrieval Conference (ISMIR), pp. 376-382, 2017.

Contact

If you are interested in technical details and applications, please contact me:

Eita Nakamura enakamura[at]sap.ist.i.kyoto-u.ac[dot]jp
Speech and Audio Processing Group, Graduate School of Informatics, Kyoto University External Site