Automatic Key Modulation Monitoring and Tonicization Detection with Hierarchical Clustering

In a wide range of tasks involving tonal music, key annotation is essential. Among these, the handling of modulation remains a blind spot in most existing approaches. As a result, music analysis often struggles to cope with highly modulating and tonally diverse material. Moreover, with the growing trend of artificial intelligence researches in music generation and style transfer, there is an increasing demand for datasets containing more precise key information. This paper proposes an unsupervised algorithm capable of rapidly and effectively detecting the key of symbolic music, while also addressing any instances of modulation or tonic deviation. We evaluate the method on diverse samples, both with and without modulation, across multiple musical styles. Comparative analysis with established algorithms demonstrates its promising performance.