Sobolev norms improve measurement of audio morphing quality
Sobolev Norms in Neural Embeddings Measure Audio Morphing Regularity
Sound
Summary
Audio morphing creates new sounds by blending two different sounds, but it's hard to measure how smooth or natural these blends are. The authors introduce a new mathematical way, called Sobolev Distances to Ideal Morphing (SDIM), to objectively judge how well audio morphing works by looking at its paths in a sound-representing space. They tested SDIM using a sound synthesizer and found it better at telling good blends from bad ones than existing methods. This helps ensure that the transitions between sounds feel more natural to listeners.
What this means in practice
- •For sound engineers: Identify and quantify smooth audio transitions in sound design using the SDIM metric to improve creative workflows.
- •For music producers: Use SDIM to evaluate and select audio morphs with natural perceptual quality when creating new musical sounds.
Authors
Théo Chasle Cauchy, Modan Tailleur, Barbara Pascal, Fanny Roche, Mathieu Lagrange
Abstract
Morphing has recently gained renewed interest with the emergence of generative models, particularly in audio and image generation. In musical sound synthesis, morphing can generate intermediate sounds between two targets, helping musicians and sound engineers explore new sounds with interesting perceptual properties. As morphing is inherently defined in perceptual terms, evaluating this task is challenging. In this work, we introduce Sobolev Distances to Ideal Morphing (SDIM), a novel objective metric to quantify the regularity of audio morphing trajectories in perceptually relevant audio embedding spaces. Leveraging a physics-based sound synthesizer, we evaluate the discriminative power of SDIM on controlled morphing trajectories with varying degrees of regularity and compare it with that of existing audio morphing metrics. Results show that, contrary to state-of-the-art metrics, the proposed metric reliably discriminates desirable trajectories from adversarial ones.