Pembuatan Musik Latar Game Yang Dinamis Menggunakan Metode Text-To-Music

Silalahi, Frederick Ivan Parulian (2025) Pembuatan Musik Latar Game Yang Dinamis Menggunakan Metode Text-To-Music. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5024201071-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Industri game modern menuntut konten yang imersif dan adaptif, salah satunya adalah musik latar yang dinamis. Namun, proses produksi musik secara konvensional memakan biaya dan waktu yang signifikan. Teknologi AI, khususnya model Text-to-Music, menawarkan solusi alternatif untuk menghasilkan musik secara cepat dan efisien berdasarkan deskripsi teks. Penelitian ini bertujuan untuk menganalisis dan membandingkan tiga model generatif audio terkemuka: MusicGen dari Meta, Stable Audio dari Stability AI, dan AudioLDM, untuk menentukan efektivitasnya dalam pembuatan musik latar game yang dinamis. Metode penelitian dilakukan melalui survei terhadap responden untuk menilai aspek musikalitas, kualitas audio, kesesuaian konteks game, dan potensi perulangan (loopability). Objek penelitian adalah klip audio yang dihasilkan dari serangkaian 5 prompt standar yang mencakup skenario game. Hasil penelitian menunjukkan bahwa Stable Audio memiliki kualitas yang bagus dan lagunya mengikuti prompt yang dikasih, tetapi melodi lagu yang dihasilkan sedikit acak-acak. MusicGen menonjol dalam hal kualitas dan lagu yang repetitif. Sementara AudioLDM menghasilkan kualitas lagu terendah dan hasilnya tidak mengikuti prompt yang dikasih. Penelitian ini menyajikan sebuah panduan bagi pengembang game dalam memilih model Text-to-Music yang paling sesuai dengan kebutuhan produksi musik latar yang dinamis.
=================================================================================================================================
The modern game industry demands immersive and adaptive content, one of which is dynamic background music. However, the conventional music production process is costly and time-consuming. AI technologies, particularly Text-to-Music models, offer an alternative solution to quickly and efficiently generate music based on text descriptions. This research aims to analyze and compare three leading audio generative models: MusicGen from Meta, Stable Audio from Stability AI, and AudioLDM, to determine their effectiveness in dynamic game background music generation. The research method was conducted through a survey of respondents to assess aspects of musicality, audio quality, game context appropriateness, and loopability. The object of research was audio clips generated from a series of 5 standardized prompts covering the game scenario. The results showed that Stable Audio had good quality and the songs followed the prompts given, but the melodies of the generated songs were a little disjointed. MusicGen stood out in terms of quality and repetitive songs. While AudioLDM produced the lowest quality songs and the results did not follow the prompts. This research presents a guide for game developers in choosing the Text-to-Music model that best suits the needs of dynamic background music production.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Text-to-Music, Musik Game Dinamis, Generative AI, MusicGen, Stable Audio, AudioLDM, Text-to-Music, Dynamic Game Music, Generative AI, MusicGen, Stable Audio, AudioLDM
Subjects:	M Music and Books on Music > M Music
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Computer Engineering > 90243-(S1) Undergraduate Thesis
Depositing User:	Frederick Ivan Parulian
Date Deposited:	31 Jul 2025 01:14
Last Modified:	31 Jul 2025 01:14
URI:	http://repository.its.ac.id/id/eprint/123300

Actions (login required)

View Item