ANIMETAB A NEW GUITAR TABLATURE DATASET OF ANIME AND GAME MUSIC Yuecheng Zhou

2025-04-27 0 0 4.88MB 8 页 10玖币

侵权投诉

ANIMETAB: A NEW GUITAR TABLATURE DATASET OF ANIME AND

GAME MUSIC

Yuecheng Zhou

Communication University of China

Yaolong Ju

Mcgill University

Lingyun Xie

Communication University of China

ABSTRACT

While guitar tablature has become a popular topic in MIR

research, there exists no such a guitar tablature dataset that

focuses on the soundtracks of anime and video games,

which have a surprisingly broad and growing audience

among the youths. In this paper, we present AnimeTAB,

a ﬁngerstyle guitar tablature dataset in MusicXML format,

which provides more high-quality guitar tablature for both

researchers and guitar players. AnimeTAB contains 412

full tracks and 547 clips, the latter are annotated with musi-

cal structures (intro, verse, chorus, and bridge). An accom-

panying analysis toolkit, TABprocessor, is included to fur-

ther facilitate its use. This includes functions for melody

and bassline extraction, key detection, and chord label-

ing, which are implemented using rule-based algorithms.

We evaluated each of these functions against a manually

annotated ground truth. Finally, as an example, we per-

formed a music and technique analysis of AnimeTAB us-

ing TABprocessor. Our data and code have been made

publicly available for composers, performers, and music

information retrieval (MIR) researchers alike.

1. INTRODUCTION

As one of the guitar score notations, tablature can be traced

back in 16th century, when it is originally known as Cifra,

meaning something written in code in Arabic [1]. Cifra

uses numbers on horizontal lines to represent fret posi-

tions on guitar ﬁngerboard, and it has gradually evolved

into today’s guitar tablature. This simple and intuitive

way of notation enables beginners to play the instrument

easily, without requiring complicated music theory knowl-

edge. Interestingly, we found an overlap between ﬁnger-

style guitar players and Japanese anime fans [2–4]. While

not a popular ﬁeld in the world-music pantheon, Japanese

pop music, anime, and video games have a surprisingly

broad and growing audience among Asia youths especially

young guitar players.

In recent years, researchers have also shown interest in

guitar music and tablature [5–11]. However, high-quality

symbolic guitar tablature is not easy to obtain, especially

for anime music, and this limitation has likely limited the

computational study of guitar tablature to date. Table 1

lists all the open-source symbolic guitar score datasets we

found [12–16]. Sarmento et al. [12] presented a dataset

of about 26,000 band scores in GuitarPro 1and encoded

1https://www.guitar-pro.com/

Figure 1. An example of staff and guitar tablature with

special tuning and capo.

format, named DadaGP. However, the encoding tool they

released is only suitable for GuitarPro 3, GuitarPro 4, and

GuitarPro 5, which have long since ceased to be used (the

latest version has been released to GuitarPro 8), and most

tracks of DadaGP are for full bands rather than individ-

ual ﬁngerstyle players. Chen et al. [15] proposed a new

dataset named EGDB, which contained transcriptions of

the electric guitar performance of 240 tablatures rendered

with different tones. Xi et al. [13] presented a guitar music

transcription dataset: GuitarSET, which including styles

of rock, bossa nova and jazz. Burlet et al. [14] compiled

two ground-truth datasets for polyphonic transcription and

guitar tablature arrangement from manual transcriptions.

Kehling et al. [16] presented a novel dataset of electric

guitar recordings with extensive annotation of note param-

eters. As shown above, none of them focus on ﬁngerstyle

arrangements for Japanese anime music or provide musical

structure annotations. To this end, we propose a new gui-

tar tablature dataset and an accompanying analysis toolkit.

Three main contributions are presented in this paper:

• We introduce AnimeTAB, a dataset of ﬁngerstyle

guitar tablature in MusicXML format, to provide

more high-quality guitar tablature for both guitar

players and researchers. It contains 412 full tracks

of guitar tablature from anime or video game music.

• We choose 202 tablature from all the 412 tracks and

segment them into 547 clips, labeling each with a

start/end bar number and one of four musical struc-

tures: intro, verse, chorus, and bridge. These anno-

tations can be used for a variety of MIR tasks, such

as musical structure segmentation, music similarity,

and automatic composition.

• We also present TABprocessor, an analysis toolkit

for AnimeTAB. This contains several rule-based al-

gorithms to identify keys, chords, melodies, and

arXiv:2210.03027v1 [cs.SD] 6 Oct 2022

Table 1. Information of open-source guitar symbolic datasets.

Name Size Type Genre Format

DadaGP [12] 26000 symbolic 739 different genres GP3/GP4/GP5

GuitarSet [13] 30 symbolic/audio rock/jazz/funk/bossa nova JAMS

ROBOTABA [14] 75 symbolic/audio ultimate guitar top 100 list musicXML/MEI

EGDB [15] 240 symbolic/audio solo/arpeggios tablature and audio

IDMT-SMT-GUITAR [16] 64 symbolic/audio 6 different genres tablature and audio

AnimeTAB(Ours) 412 full/547 clips symbolic anime musicXML

basslines. These generated annotations are computa-

tionally inexpensive, and our experiments show that

they maintain a high degree of accuracy. Addition-

ally, we use TABprocessor to present different music

statistics of AnimeTAB, including frequency distri-

butions of keys, chords, ﬁngering, and guitar tech-

niques.

This paper is structured as follows. Section 2 gives an

overview of some published guitar symbolic datasets, sec-

tion 3 introduces the basic information of our AnimeTAB

dataset; section 4 describes the principles and functions of

TABprocessor; section 5 details the use of TABprocessor

to analyze our dataset for an example, including analysis

of notes, harmonies and techniques. Conclusions are pre-

sented in section 6. We released all the tracks, demos, and

source code in [anonymized], where all the test results can

also be found.

2. ANIMETAB DATASET

AnimeTAB is a structure-labeled guitar tablature dataset

of anime and video game music, which can be used as

symbolic data for music generation, song classiﬁcation and

recommendation, automatic accompaniment, phrase sepa-

ration, and other MIR tasks while satisfying the daily prac-

tice needs of anime ﬁngerstyle guitar players. The dataset

consists of 412 scores and 547 labeled clips from anime

episodes and video game theme songs from the 1990s to

present. All tracks are collected from open-source scores

on the Internet, most of which are written and voluntar-

ily shared by the guitar players in anime fans community.

Anime lovers always have an exceptional enthusiasm to

share their preferences and productions, but the overall

quality of their sharing are sometimes unstable. We man-

ually checked every score to ensure the quality and con-

sistency, and corrected the wrong note, illegal notations,

inappropriate format, and other mistakes. Metamessages

like Due to the requirements of different tasks, the Ani-

meTAB dataset is divided into full scores, score clips, and

encoded scores, which will each be introduced in detail.

2.1 Part 1:Full Scores

The ﬁrst part is a set of about 412 full scores. These are

derived from GuitarPro’s native GTP format and were ex-

ported to MusicXML format using GuitarPro 7. We faith-

fully recorded and retained all the aspects of these tracks

including special tunings, capo positions, and time signa-

tures for special requirements in the future. Since many of

the notations are speciﬁc for guitar techniques in GuitarPro

software, to reproduce the complete track information, we

do not recommend using other score editing software like

Musescore to open them.

2.2 Part 2:Score Clips

Instead of entire tracks, speciﬁc music phrases like verses

or chorus are more suitable in some special applications,

for example, soundtracks for short videos or game scenes.

The authors and another ﬁve students with background in

music have selected 202 from all 412 scores, extracting

speciﬁc music structures–intro, verse, bridge, and chorus–

and labeled them with their structure types and start-

ing/ending bar numbers. There are 4-16 measures in each

of these slices, meta messages from the origin songs were

retained. For the sake of data consistency in subsequent

analysis, we set the capo 2frets to zero for all clips while

maintaining the ﬁngering. More speciﬁc information is

given in Table 2. Due to our careful selection, these score

clips can also serve as practice tracks for guitar players.

Structure Number Average bar number Average note number

Intro 82 8.07 51.86

Verse 203 13.07 88.54

Chorus 184 13.01 86.11

Bridge 78 9.08 60.46

Table 2. Information of musical structure clips.

2.3 Part 3:Preprocessed Scores

Well-known natural language processing (NLP) models

used in MIR ﬁelds, such as Transformer [17], BERT

[18],and GPT-2 [19], can regard encoded musical tokens

as actual texts for further processing. Moreover, Encoded

data can take up less space and circumvent copyright is-

sues. Precisely for these reasons, We extracted the pitch,

beats, and ﬁngering information from the 547 clips and de-

signed this into the following format.

•Pitch. In every note in MusicXML ﬁles, the step,

octave, and alter nodes under the pitch node

2A device used on the neck of a stringed (typically fretted) instrument

to transpose and shorten the playable length of the strings—hence raising

the pitch

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

ANIMETAB:ANEWGUITARTABLATUREDATASETOFANIMEANDGAMEMUSICYuechengZhouCommunicationUniversityofChinaYaolongJuMcgillUniversityLingyunXieCommunicationUniversityofChinaABSTRACTWhileguitartablaturehasbecomeapopulartopicinMIRresearch,thereexistsnosuchaguitartablaturedatasetthatfocusesonthesoundtracksofanimea...

展开>> 收起<<

ANIMETAB A NEW GUITAR TABLATURE DATASET OF ANIME AND GAME MUSIC Yuecheng Zhou.pdf

共8页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

ANIMETAB A NEW GUITAR TABLATURE DATASET OF ANIME AND GAME MUSIC Yuecheng Zhou

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: