Implementation of a Music Generation System Through Automatic Text Prompt Generation: Focusing on Emotion Analysis Through Music Analysis and Lyrics Analysis
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Current AI-based music generation models and research primarily focus on manual text-based music generation. This paper proposes a music generation system that automates text prompts to enhance user convenience and streamline the creative process. The study involves building a dataset that includes genre, artist, and album information by analyzing and processing music data collected from audio files and extracting keywords from lyrics. The lyrics data are tokenized using the KoNLPy natural language processing library in Python, and key terms are extracted through TF-IDF vectorization. Additionally, the study suggests a method for automatically generating text prompts using MFCC, tempo, and other feature data to predict emotions through a model that combines CNN and ChatGPT. These automatically generated text prompts are then input into the MusicGen model to automatically create new music that reflects the user's emotional state and musical preferences. The findings of this study are expected to contribute to the field of music data analysis and generation.
Abstract
Current AI-based music generation models and research primarily focus on manual text-based music generation. This paper proposes a music generation system that automates text prompts to enhance user convenience and streamline the creative process. The study involves building a dataset that includes genre, artist, and album information by analyzing and processing music data collected from audio files and extracting keywords from lyrics. The lyrics data are tokenized using the KoNLPy natural language processing library in Python, and key terms are extracted through TF-IDF vectorization. Additionally, the study suggests a method for automatically generating text prompts using MFCC, tempo, and other feature data to predict emotions through a model that combines CNN and ChatGPT. These automatically generated text prompts are then input into the MusicGen model to automatically create new music that reflects the user's emotional state and musical preferences. The findings of this study are expected to contribute to the field of music data analysis and generation.
참고문헌 (12)
- Agostinelli, A., "MusicLM: Generating High-Quality Music from Text Descriptions", Proceedings of the 2023International Conference on Machine Learning (ICML), 6, pp.1234~1245 (2023)
- Baltrusaitis, T., "Multimodal Machine Learning: A Survey and Taxonomy", IEEE Transactions on Pattern Analysis and Machine Intelligence, 41(2), pp.423~443 (2018)
- Choi, K., "Improving Music Genre Classification Performance using GAN", Journal of Music Technology Research, 12(3), pp.145~158 (2023)
- Copet, J., "MUSICGEN : A Unified Language Model for Generating Music with Multiple Streams of Compressed Discrete Representations", International Journal of Music AI Research, 8(1), pp.1~12 (2023)
- Goodfellow, I., "Deep Learning" (2014), MIT Press
- Yang, Z., "Diffsound : A Non-Autoregressive Model for Generating Melody-Based Sounds", Proceedings of the 2023 Conference on Neural Information Processing Systems (NeurIPS), 1, pp.1~10 (2023)
- Yoon, H., "Research on Emotional Music Generation Using Deep Learning and Russell Model-Based Emotional Vectors", Proceedings of the Korean Society of Communications Conference, 2024(1), pp.1440~ (2024)
- Kim, J., "Development and Evaluation of a Dataset for Automatic Music Emotion and Genre Classification", Journal of Next Generation Computing, 19(3), pp.74~82 (2023)
- Park, B., "Analysis of Domestic and International Research Trends in the Music Field Related to Artificial Intelligence", Journal of Music Pedagogy Research, 25(1), pp.1~22 (2024)
- Park, W., "A Study on Emotion Classification Model for Parting Lyrics Using LSTM", Journal of Korean Emotion Analysis, 15(2), pp.123~135 (2020)
- Lee, J., "A Knowledge Graph-Based Music Recommendation System Using Lyric Keyword Extraction", Journal of the Korean Data & Information Science Society, 33(6), pp.937~949 (2022)
- "Term Frequency-Inverse Document Frequency (TF-IDF)" (2024)[URL]