notesum.ai

Published at November 20

Song Form-aware Full-Song Text-to-Lyrics Generation with Multi-Level Granularity Syllable Count Control

cs.CL

cs.AI

Released Date: November 20, 2024

Authors: Yunkee Chae¹, Eunsik Shin², Hwang Suntae², Seungryeol Paik², Kyogu Lee³

Aff.: ¹IPAI, Seoul National University; ²Department of Intelligence and Information, Seoul National University; ³IPAI, Department of Intelligence and Information, AIIS, Seoul National University

Arxiv: http://arxiv.org/abs/2411.13100v1

Model

SCD

\downarrow

SCErr (%)

\downarrow

PPL

\downarrow

BERT-S

\uparrow

Full

Para.

Line

Phrase

Word

Full

Para.

Line

Phrase

Word

Full

Para.

Line

Phrase

Word

Front-P

0.026^∗

0.068^∗

0.013^∗

0.021^∗

0.024^∗

10.313^∗

84.856^∗

11.711^∗

6.140^∗

3.618^∗

17.069

31.259

28.442

16.793

11.271

0.735^∗

Front-S

0.006^∗

0.047^∗

0.004^∗

0.002

5.501^∗

75.781^∗

3.998^∗

0.759

0.252

19.036

35.613

31.321

18.377

12.218

0.736^∗

Both-P

0.016^∗

0.043^∗

0.009^∗

0.012^∗

0.015^∗

8.247^∗

79.516^∗

8.596^∗

3.944^∗

2.178^∗

17.048

31.652

28.421

16.831

11.259

0.737^∗

Both-S

0.005^∗

0.041^∗

0.003^∗

0.002

0.001

5.404^∗

74.847^∗

3.501^∗

0.778

0.244

19.453

36.423

31.972

18.847

12.410

0.737^∗

Back-P

0.014^∗

0.037^∗

0.009^∗

0.011^∗

0.014^∗

7.516^∗

76.813^∗

7.606^∗

3.194^∗

2.041^∗

17.776

34.667

30.344

17.633

11.847

0.739^∗

\hdashlineBack-S

0.003

0.027

0.003

0.002

5.025

68.777

2.851

0.792

0.266

19.883

39.189

33.867

19.422

12.802

0.740