Abstract
The challenges of exploring the latent spaces of deeplearning audio models often lead artists to rely on chance,
randomness, and combinatorial approaches, making it difficult to steer these models toward musically meaningful
outcomes. In this paper, we explore how Principal Component Analysis (PCA) applied to pre-encoded RAVE (Re-
altime Audio Variational Autoencoder) latent representations can provide a more controlled and curated approach
to navigating these high-dimensional spaces. By restricting exploration to selected regions of the latent space, mu-
sicians gain clearer pathways to achieving specific sonic goals. Although t-SNE and UMAP effectively preserve in-
tricate local structures, we show how the linearity, computational efficiency, and interpretability of PCA offer dis-
tinct advantages for real-time applications. In addition, we introduce a graphical user interface (GUI) and a sensor sys-
tem for manipulating ’timbral vectors’ derived from PCA components, providing an intuitive tool for identifying, re-
fining, and shaping sonic transformations. To evaluate the effectiveness of PCA, we systematically compare its per-
formance with t-SNE and UMAP, highlighting the trade-offs among these methods.