Visualization method of sound effect retrieval based on UMAP
Yang Jiale, Ying Zhang · 2020 IEEE 4th Information Technology, Networking, Electronic and Automation Control Conference (ITNEC) · 2020
In the scatter plot-based sound effect retrieval system, a dimensionality reduction algorithm is used to convert a high-dimensional sound effect data set into a low-dimensional data set, and visually distribute it on a plane. Traditional systems often use a non-linear dimensionality reduction algorithm t-distributed Stochastic Neighbor Embedding (t-SNE), but its dimensionality reduction speed is insufficient, and it pays more attention to the local structure of data distribution, and does not explicitly retain the global structure. This paper proposes to use Uniform Manifold Approximation and Projection (UMAP) algorithm to improve these problems, and compares the efficiency and stability of these two algorithms by visualizing the sound effect data set. Experiments show that under the premise that the distribution structure relationship between sound effects can be well described, UMAP takes about 5 times faster than t-SNE, and UMAP also performs better than t-SNE in terms of stability. Therefore, UMAP is a better choice of dimensionality reduction algorithm when using the scatter plot to visualize the sound effects library.