LSTM 特徴抽出と楽譜分析による特定のピアニストの演奏スタイルを模倣する自動演奏に関する研究

• Item type: SIG Technical Reports(1)
• 公開日: 2024-02-22
• タイトル: LSTM 特徴抽出と楽譜分析による特定のピアニストの演奏スタイルを模倣する自動演奏に関する研究
• タイトル / 言語: en
• タイトル: A Study on Automatic Performance for Emulating the Playing Style of a Specific Pianist using Feature Extraction with LSTM and Score Analysis
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: SP2
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: 日本大学大学院理工学研究科情報科学専攻
• 著者所属: 日本大学大学院理工学研究科情報科学専攻
• 著者所属(英): en
• 著者所属(英): Department of Computer Engineering, College of Science and Technology, Nihon University
• 著者所属(英): en
• 著者所属(英): Department of Computer Engineering, College of Science and Technology, Nihon University
• 著者名: 李, 森浩; 松野, 裕
• 著者名(英): Senhao, Li; Yutaka, Matsuno
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Classical music continues to captivate audiences worldwide, and advancements in automatic piano playing technologies and the Internet allows people to enjoy professional piano performances at home. However, the labor-intensive process of recording tracks on sensor-equipped pianos and the limited repertoire available present challenges. Thus, in this paper, we propose a neural network–based method to generate automatic piano performances. The proposed method utilizes long short-term memory (LSTM) networks, designed for the specific pianist, and contributes to the field of music notation learning. We employ LSTM networks rather than a recurrent neural network to decipher and encapsulate the intricate temporal dependencies characteristic of piano performances. We conducted training loss evaluation and pianist perception assessment. Experimental outcomes demonstrated predictive accuracy with R-squared (R2) values between 0.4 and 0.6. Blind listening tests revealed the model’s effectiveness in generating performances that are comparable to those of skilled pianists.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Classical music continues to captivate audiences worldwide, and advancements in automatic piano playing technologies and the Internet allows people to enjoy professional piano performances at home. However, the labor-intensive process of recording tracks on sensor-equipped pianos and the limited repertoire available present challenges. Thus, in this paper, we propose a neural network–based method to generate automatic piano performances. The proposed method utilizes long short-term memory (LSTM) networks, designed for the specific pianist, and contributes to the field of music notation learning. We employ LSTM networks rather than a recurrent neural network to decipher and encapsulate the intricate temporal dependencies characteristic of piano performances. We conducted training loss evaluation and pianist perception assessment. Experimental outcomes demonstrated predictive accuracy with R-squared (R2) values between 0.4 and 0.6. Blind listening tests revealed the model’s effectiveness in generating performances that are comparable to those of skilled pianists.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AN10442647
• 書誌情報: 研究報告音声言語情報処理（SLP）; 巻 2024-SLP-151, 号 30, p. 1-6, 発行日 2024-02-22
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8663
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会