~Proceedings ICMCISMCI2014 14-20 September 2014, Athens, Greece
Unso-n-er
Division Copy-and-paste
~Assigned singing are
different between sections
[drag
v_ and drop vvvv. vvavvvv
The song can
be divided into
sections Da
section ~and drop
Adjustment
Border of section signe singing an t air
can be adjusted voume become same
by drag-and-drop
CutpneBtrnatner usr Lste
InterfNceber f
Cut Ska/iltreg Bttngerwith sDiie h sn'it
Butn cusisfaursan etdsaE eon usningriTrogs.
A..........................................s..........p.....s....ng.......n...g.....a ndng n.... s t.....m...k....k e a\., m p\ \.. n ts
Focuse Add Rem. ove >. \.............. F
Figure 2. Overview of Unisoner and the interaction between user and Unisoner.
derivative singings at the same time allows listeners to
notice differences in singing style, vocal timbre, etc.
Discover favorite singers Derivative choruses give listeners a chance to discover singers they like from the
derivative singings used in the choruses. Some creators
of derivative choruses mash up derivative singings to
highlight their favorite singers.
Creators of derivative choruses can also enjoy the derivative creation.
It was, however, not easy to create derivative choruses.
First, it is necessary to extract singing voices from derivative singings by suppressing their karaoke accompaniments. Second, since the extracted singing voices are not
temporally aligned, it is time-consuming to synchronize
them with a karaoke accompaniment. Third, creators must
use a waveform-editing tool, such as Digital Audio Workstation (DAW), to manually cut and paste fragments of
singing voices from different singers.
We therefore propose an easy-to-use interface, Unisoner,
that enables end users without musical expertise to create
derivative choruses. Unisoner overcomes the difficulties
described above by automating the synchronization tasks
necessary for derivative choruses and providing intuitive
mixing functions. This work forms part of the emerging field of creative Music Information Retrieval (MIR),
where MIR techniques are used for creative purposes. In
this field, there have been interfaces for creating mashup music by connecting loops corresponding to musical
pieces [2], creating mash-up music automatically [3], and
creating mash-up dance videos [4], yet no interface for
derivative choruses has been reported.
In addition to Unisoner, we also propose a singing training interface that leverages various derivative singings.
Since amateur singers have difficulty improving their
singing skill, singing training interfaces, such as an interface to analyze a singer's voice alone [5] and an interface
to compare two singing voices [6], have been proposed.
Our training interface allows a user to compare his or her
singing with a wide variety of derivative singings by visualizing them. The user can choose a favorite existing
singing by using metadata such as the number of views
on a video sharing service, and compare the fundamental
frequency (FO) of the chosen singing with the Fo of the
user's singing so that the user can sing more like the favorite singing.
Demonstration videos of Unisoner are available at
http://mmlab.cs.tsukuba.ac.jp/%7etsuzuki/icmcsmc14.
2. UNISONER: INTERACTIVE DERIVATIVE
CHORUS CREATION INTERFACE
Unisoner enables users to create derivative choruses easily,
and allows for the simultaneous listening of various derivative singings and derivative choruses. We assume audio
signals have been extracted from a set of desired videos on
YouTube or Niconico. We call the resulting audio signal
the accompanied singing, and the vocal audio signal after
vocal extraction (see Section 3) the suppressed singing.
2.1 Interface of Unisoner
Figure 2 shows an overview of Unisoner. Creators of
derivative choruses using conventional methods (e.g.,
waveform-editing tools) had to work hard to make derivative choruses, such as by cutting and pasting fragments of
suppressed singings or adjusting the volume of each suppressed singing. Unisoner provides users with an easy-touse interface to overcome these difficulties.
Unisoner displays each suppressed singing as an icon that
represents each singer (the singer icon). A user can assign each singing to phrases and adjust volume simply by
dragging and dropping singer icons. Moreover, musicsynchronized lyrics, given in advance, enable the user to
assign each singing to certain phrases easily.
791 -
0