Transformer-Based Seq2Seq Model for Chord Progression Generation

Machine learning is widely used in various practical applications with deep models demonstrating advantages handling huge data. Treating music as a special language and using to accomplish melody recognition, generation, analysis has proven feasible. In certain music-related research, recurrent neural networks have been replaced transformers. This achieved significant results. traditional approaches networks, input sequences are limited length. paper proposes method generate chord progressions for melodies transformer-based sequence-to-sequence model, which divided into pre-trained encoder decoder. A extracts contextual information from melodies, whereas decoder uses this produce chords asynchronously finally outputs progressions. The proposed addresses length limitation issues while considering the harmony between melodies. Chord can be generated composition applications. Evaluation experiments conducted three baseline models. included bidirectional long short-term memory (BLSTM), representation transformers (BERT), generative transformer (GPT2). outperformed Hits@k (k = 1) by 25.89, 1.54, 2.13 %, respectively.