Interpretable Spectral Parameters and Interface for Music Timbre Design

Spectral music and spectral modeling have been explored for decades, during which many models have been established for music information retrieval (MIR) and generative purposes. However, in synthesis scenarios, most spectral representations, if not all of them, suffer from a trade-off between the fidelity of reconstruction and the controllability of their parameters. This trade-off leads to a strong dependency on reference sounds and thus limits the exploration of musical timbre with spectral approaches. This work defines a compact set of control parameters on the morphology of the spectrogram with good interpretability. Based on this representation, a parameter interface and a graphical interface are designed, which together enable real-world users to perform spectral sound design from scratch. The validity of the representation is verified with an instrument classification experiment and several synthesis tasks. Finally, this work initiates a discussion on the potential and significance of spectral synthesis for both acoustic research and music creation. This work was completed in 2020 as part of the author's master's thesis. It has not been previously published and is posted here for archival purposes.