Does it Chug? Towards a Data-Driven Understanding of Guitar Tone Description

Pratik Sutar, Jason Naradowsky, Yusuke MiyaoarXiv 2024 · DOI · arXiv

Introduction

In the world of guitar performance, communication often relies on a rich, subjective vocabulary. Terms like "creamy," "brown sound," or "chug" are common parlance among musicians, yet they remain notoriously difficult to quantify scientifically. This ambiguity creates a significant hurdle for developing AI-driven music tools that can interpret and act upon musician-led descriptions. Researchers from the University of Tokyo and Square-Enix have addressed this gap by creating a data-driven dataset that maps subjective adjectives to specific guitar timbres.

Methodology: Isolating Timbre

To move beyond traditional psychoacoustic theories, the researchers developed a controlled three-step approach to isolate timbre from musical performance:

  1. Lexicon Building: A crawl of online guitar forums and articles identified 110 commonly used adjectives.
  2. Audio Generation: Twelve "Direct Input" (DI) tracks—clean guitar signals—were recorded across various playing styles. These were processed through 80 different digital effects presets using the Helix Native plugin, creating 960 unique audio samples where the only variable was the signal chain (timbre).
  3. Expert Crowdsourcing: 38 experienced guitarists (most with over 10 years of experience) performed pairwise comparisons and labeling tasks.
  4. Unification: A graph-based algorithm unified these annotations into a single "relevance score" for each clip-adjective pair.

The "Does it Chug?" Dataset

The resulting "Does it Chug?" dataset provides a benchmark for understanding guitar-specific descriptors. It includes 110 adjectives, 960 audio clips, and over 2,000 expert annotations.

Frequency of the most common timbral adjectives used by experts.

Analysis of the label frequency shows a clear emphasis on rock and metal aesthetics, with "Distorted," "Thick," and "Crunchy" appearing most frequently. Despite the subjective nature of sound, the experts showed high agreement; in cases where multiple experts rated the same pair of clips, they only disagreed on the ranking in 1 out of 6 instances.

Key Insights: The Brightness Paradox and Semantic Clusters

One of the study's most significant findings challenges the traditional link between "brightness" and "spectral centroid" (high-frequency energy). The data revealed instances where experts labeled a clip as "less bright" despite it having a higher spectral centroid, suggesting that fundamental frequency and harmonic-to-noise ratios play a more critical role in human perception than previously assumed.

Furthermore, the study identified "semantic clusters"—groups of adjectives that naturally clump together. For instance, a "full" tone is statistically likely to be described as "distorted," "thick," and "dark." This anchoring helps define ambiguous terms through their relationship with more established descriptors.

Mapping Adjectives to Presets

The research successfully identified which specific amplifier and effect models correlate most strongly with certain descriptors. This mapping provides a concrete link between subjective language and technical signal processing.

Mapping of specific adjectives to their most relevant Helix Native presets.

For example, the "Line 6 Litigator" was identified as the quintessential "Warm" preset, while the Marshall-style "Brit 2204" was the primary reference for "Distorted" tones. This data highlights that certain adjectives only make sense in specific musical contexts; for instance, the term "chug" is inextricably linked to percussive, palm-muted rhythm playing and cannot be applied to clean arpeggios.

Conclusion and Future Directions

While the study is limited by a relatively small expert pool and a slight genre bias toward rock and metal, it represents a major step toward "prompt-based" music production. By quantifying the "unquantifiable," this research paves the way for AI tools that can understand and execute complex timbral requests, such as "make this guitar sound more liquid and boutique."

Summary updated

Abstract

Natural language is commonly used to describe instrument timbre, such as a "warm" or "heavy" sound. As these descriptors are based on human perception, there can be disagreement over which acoustic features correspond to a given adjective. In this work, we pursue a data-driven approach to further our understanding of such adjectives in the context of guitar tone. Our main contribution is a dataset of timbre adjectives, constructed by processing single clips of instrument audio to produce varied timbres through adjustments in EQ and effects such as distortion. Adjective annotations are obtained for each clip by crowdsourcing experts to complete a pairwise comparison and a labeling task. We examine the dataset and reveal correlations between adjective ratings and highlight instances where the data contradicts prevailing theories on spectral features and timbral adjectives, suggesting a need for a more nuanced, data-driven understanding of timbre.
BibTeX
@misc{sutar2024does,
  author = {Pratik Sutar and Jason Naradowsky and Yusuke Miyao},
  title = {{Does it Chug? Towards a Data-Driven Understanding of Guitar Tone Description}},
  year = {2024},
  eprint = {2412.11769},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2412.11769},
}
Figure