Sound of the Mix: Identifying Mixing Engineers With Deep Learning

Advances in music production technology at both hardware and software levels have led to a substantial increase in the volume of music released annually, motivating the development of automated music categorization systems based on attributes such as artist, genre, and mood. In contrast, automated categorization based on mixing engineers remains largely unexplored, despite their critical role in shaping a song’s final sonic identity. Mixing engineers translate artistic intent into a cohesive auditory experience through nuanced control of frequency balance, dynamics, spatial depth, and stereo imaging. This paper proposes a deep learning–based framework for identifying mixing engineers by analyzing distinctive sonic characteristics embedded in stereo audio recordings. The approach processes stereo channels individually and isolates key interchannel components to capture subtle mixing-related cues. The extracted representations are visualized and processed using dedicated convolutional neural networks, whose outputs are fused to enable multidimensional sonic analysis. Experimental evaluation on a dataset of 60 songs, predominantly spanning the modern rock and metal spectrum, achieved a maximum classification accuracy of 92.86%. The recordings were produced by six mixing engineers, with artist-related similarities controlled for during evaluation, demonstrating the effectiveness of the proposed method.