The Algorithmic Flattening of Sound: Computational Evidence and Justice Implications of AI Music Homogenization

May combine two claims.

Suno outputs show higher within-genre spread than human tracks (variance ratio 1.667; 83% of feature×genre cells show AI > human; Cohen's d = +0.646).

“Suno outputs show higher within-genre spread than human tracks (variance ratio 1.667; 83% of feature×genre cells show AI > human; Cohen's d = +0.646),with normalized pairwise distances exceeding 1.0 in three of four genres (range: 1.004-1.582).”

• AssumptionRests on an assumption: The human reference corpora (100 tracks per genre from Spotify playlists) adequately represent the acoustic diversity of each genre.

May combine two claims.

Lyria tracks tempo moderately and onset density in some genres but shows limited fidelity for harmonic ratio and self-similarity.

“Lyria tracks tempo moderately and onset density in some genres but shows limited fidelity for harmonic ratio and self-similarity (Figure 1).”

May combine two claims.

Suno and Lyria do not converge on a common acoustic profile; they are more acoustically distant from each other than two random human subsamples would typically be.

“The two systems do not converge on a common acoustic profile and are more acoustically distant from each other than two random human subsamples would typically be.”

• Also stated in: § RQ3: The Two Systems Do Not Converge on a Common Sound · plain

  “The two systems are therefore more acoustically distant from each other than two random human subsets would typically be, across every genre (Figure 4).”

• AssumptionRests on an assumption: The human reference corpora (100 tracks per genre from Spotify playlists) adequately represent the acoustic diversity of each genre.

May combine two claims.

The aggregate variance ratio (AI/Human) for Lyria is 0.839, with 58% of feature×genre cells showing AI variance below human.

“The aggregate variance ratio (AI/Human) is 0.839, meaning AI output variance is on average 84% of human variance, with 58% of feature×genre cells showing AI variance below human.”

• AssumptionRests on an assumption: The human reference corpora (100 tracks per genre from Spotify playlists) adequately represent the acoustic diversity of each genre.

May combine two claims.

Neither Suno nor Lyria follows user prompts faithfully.

“Neither system follows user prompts faithfully, indicating that the observed patterns reflect learned priors rather than prompt constraints.”

• Also stated in: § Prompt Fidelity Is Low for Both Systems · plain

  “Suno shows near-zero prompt fidelity across all features (max |r| = 0.26), meaning its outputs bear almost no relationship to the acoustic properties specified in the prompt.”

May combine two claims.

Suno collapses acoustic distinctions between genres without compressing within-genre spread.

“Suno collapses the acoustic distinctions between genres without compressing within-genre spread.”

• AssumptionRests on an assumption: The human reference corpora (100 tracks per genre from Spotify playlists) adequately represent the acoustic diversity of each genre.

May combine two claims.

Suno's genre separation ratio falls 36% below the human baseline, indicating that genre boundaries collapse in its output space; Lyria matches human genre separation (+1%).

“Suno's separation ratio falls 36% below the human baseline, indicating that genre boundaries collapse in its output space.Lyria matches human genre separation (+1%).”

May combine two claims.

All convergence ratios exceed 1.0, ranging from 2.70 (Heavy Metal) to 4.64 (Afrobeats), with Dance Pop at 4.38 and K-pop at 4.47.

“All ratios exceed 1.0, ranging from 2.70 (Heavy Metal) to 4.64 (Afrobeats), with Dance Pop at 4.38 and K-pop at 4.47.”

• AssumptionRests on an assumption: The human reference corpora (100 tracks per genre from Spotify playlists) adequately represent the acoustic diversity of each genre.

May combine two claims.

Lyria's genre separation ratio is 0.676 (+1% relative to human), indicating Lyria's within-genre compression does not distort overall genre structure.

“Lyria is nearly identical at 0.676 (+1%), indicating that Lyria's within-genre compression does not distort the overall genre structure.”

• AssumptionRests on an assumption: The human reference corpora (100 tracks per genre from Spotify playlists) adequately represent the acoustic diversity of each genre.

May combine two claims.

Suno's title diversity within individual calls is low, and the vocabulary of titles is narrow across the full 100-track set for each genre.

“Overall, diversity within individual calls is low, and the vocabulary of titles is narrow across the full 100-track set for each genre (Appendix Figure 6).”

May combine two claims.

The two systems exhibit structurally distinct patterns: Lyria reduces within-genre acoustic diversity, while Suno collapses the acoustic distinctions between genres without compressing within-genre spread.

“The two systems exhibit structurally distinct patterns: Lyria reduces withingenre acoustic diversity, while Suno collapses the acoustic distinctions between genres without compressing withingenre spread.”

May combine two claims.

The three highest-importance features for AI-vs-human discriminability are MFCC ∆ 2 mean (0.144), MFCC 0 (0.117), and IOI mean (0.051), with timbral dynamics and rhythmic regularity together accounting for the majority of discriminative signal.

“The three highest-importance features are MFCC ∆ 2 mean (timbral envelope acceleration, importance = 0.144), MFCC 0 (overall energy envelope, 0.117), and IOI mean (average time between note onsets, 0.051), with timbral dynamics and rhythmic regularity together accounting for the majority of discriminative signal (Figure 5).”

• AssumptionRests on an assumption: The human reference corpora (100 tracks per genre from Spotify playlists) adequately represent the acoustic diversity of each genre.

May combine two claims.

Lyria shows moderate tempo tracking (r = 0.30-0.57) but limited fidelity elsewhere.

“Lyria shows moderate tempo tracking (r = 0.30-0.57)but limited fidelity elsewhere.”

May combine two claims.

Discriminability holds when restricting to Afrobeats (AUC = 0.980 ± 0.011), with IOI features gaining relative importance and vocal-sensitive MFCC features dropping from 64.5% to 38.0% of top-10 importance, confirming that rhythmic regularity rather than vocal content drives the separation.

“Discriminability holds: AUC remained 0.980 ± 0.011, with IOI features gaining relative importance while vocal-sensitive MFCC features dropped from 64.5% to 38.0% of top-10 importance, confirming that rhythmic regularity rather than vocal content drives the separation.”

• AssumptionRests on an assumption: The human reference corpora (100 tracks per genre from Spotify playlists) adequately represent the acoustic diversity of each genre.

May combine two claims.

Lyria consistently falls at or below the human baseline for within-genre acoustic diversity; Suno exceeds it in three of four genres, with Afrobeats showing the largest deviation (1.58×).

“Lyria consistently falls at or below the human baseline (blue dashed line); Suno exceeds it in three of four genres, with Afrobeats showing the largest deviation (1.58×).”

May combine two claims.

Genre separation ratio: Suno collapses genre boundaries (-36% vs. human); Lyria preserves them.

“Suno collapses genre boundaries (-36% vs. human); Lyria preserves them.”

May combine two claims.

Within-genre mean pairwise distance ratios: Suno exceeds 1.0 in three of four genres; Lyria consistently falls near or below 1.0.

“Lyria consistently falls near or below 1.0; Suno exceeds 1.0 in three of four genres.”

Probably not a substantial claim: may not pass S4 (based).

The concerns are especially relevant for styles that are culturally underrepresented in training data, where a system may reproduce a genre in a simplified and standardized form that is easier to scale.

“The concerns are especially relevant for styles that are culturally underrepresented in training data, where a system may reproduce a genre in a simplified and standardized form that is easier to scale.”

May combine two claims.

The MIR feature set reflects dominant Western production norms and may underrepresent musically salient distinctions in non-Western traditions, a limitation particularly relevant for Afrobeats and K-pop.

“Finally, the MIR feature set upon which we base our analyses, though widely used, reflects dominant Western production norms and may underrepresent musically salient distinctions in non-Western traditions (Peeters et al. 2025)-a limitation that is particularly relevant for Afrobeats and Kpop, the genres where cultural stakes are highest and where our audit is most justice-relevant.”

• cites Peeters et al. 2025Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Music appears headed the same way as the writers' strike, through settlements that may protect major labels more than working musicians.

“Music appears headed the same way, through settlements that may protect major labels more than working musicians.”

May combine two claims.

The outputs are not simple copies, but neither are they acoustically detached from the human musical domains on which these systems appear to rely.

“The outputs are not simple copies, but neither are they acoustically detached from the human musical domains on which these systems appear to rely.”

Classified as probably not a substantial claim. May combine two claims.

The results characterize output behavior under realistic use conditions rather than the upper bound of what more sophisticated prompting might achieve; prompt fidelity is limited for both systems, and expert prompting could partially alter the patterns observed.

“The results characterize output behavior under realistic use conditions rather than the upper bound of what more sophisticated prompting might achieve; prompt fidelity is limited for both systems, and expert prompting could partially alter the patterns observed.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

In the future, AI systems may push genres toward the patterns they most reliably generate, while human producers may be pressured to make music that fits those patterns in order to remain visible.

“In the future, AI systems may push genres toward the patterns they most reliably generate, while human producers may be pressured to make music that fits those patterns in order to remain visible.”

May combine two claims.

The results reveal two distinct homogenization patterns: Lyria compresses tracks toward a genre center; Suno compresses genre centers toward each other.

“The results reveal two distinct homogenization patterns.Lyria compresses tracks toward a genre center; Suno compresses genre centers toward each other.”

May combine two claims.

Streaming platforms rely on audio descriptors to group tracks, populate genre playlists, and optimize for engagement, a process Prey (2020) describes as curatorial power.

“Streaming platforms rely on audio descriptors to group tracks, populate genre playlists, and optimize for engagement, a process Prey (2020) describes as curatorial power”

• cites Prey 2020Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S4 (based).

The null-prompt baseline condition isolates system-level priors from prompt engineering effects.

“introduces a null-prompt baseline condition that isolates systemlevel priors from prompt engineering effects;”

May combine two claims.

The commercial pressure is to reduce cost by producing music that is functionally adequate rather than distinctive, and nothing in the patterns our audit documents suggests those pressures are working in the other direction.

“The commercial pressure is to reduce cost by producing music that is functionally adequate rather than distinctive, and nothing in the patterns our audit documents suggests those pressures are working in the other direction.”

• AssumptionRests on an assumption: The commercial pressure is to reduce cost by producing music that is functionally adequate rather than distinctive.

Probably not a substantial claim: may not pass S4 (based).

We address the vocal/instrumental confound by restricting the classifier to Afrobeats, the only genre with a fully instrumental human corpus.

“We address this directly by restricting the classifier to Afrobeats, the only genre with a fully instrumental human corpus”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Findings do not estimate population prevalence across all generative music models, genres, or time periods, nor identify internal causes of reduced variation.

“Findings do not estimate population prevalence across all generative music models, genres, or time periods, nor do they identify internal causes of reduced variation.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

The homogenization patterns documented here carry implications for what forms of musical knowledge remain legible once generation is automated, and which communities bear the cost.

“The homogenization patterns documented here carry implications for what forms of musical knowledge remain legible once generation is automated, and which communities bear the cost.”

Classified as probably not a substantial claim. May combine two claims.

These findings reflect how most users encounter these systems, since both platforms are designed and marketed for use without musical training or expertise.

“These findings nonetheless reflect how most users encounter these systems, since both platforms are designed and marketed for use without musical training or expertise.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

Both homogenization patterns become consequential once generated music enters the platform infrastructure that shapes what listeners hear, how genre definitions stabilize, and what kinds of music production are economically viable.

“Both patterns become consequential once generated music enters the platform infrastructure that shapes what listeners hear, how genre definitions stabilize, and what kinds of music production are economically viable.”

• AssumptionRests on an assumption: Platform infrastructure shapes what listeners hear, how genre definitions stabilize, and what kinds of music production are economically viable.

May combine two claims.

The scale of AI-generated supply alone makes platform-level influence plausible, even if upload volume does not tell us how often such tracks are actually heard.

“the scale of supply alone makes this a plausible site of platform-level influence, even if upload volume does not tell us how often such tracks are actually heard.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). May combine two claims.

Broader implications regarding platform dynamics, cultural homogenization, and economic impacts remain interpretive and require further investigation.

“however, broader implications regarding platform dynamics, cultural homogenization, and economic impacts remain interpretive and require further investigation.”

May combine two claims.

Access to a finished audio file and access to musical craft are two different things, with musical production being a developmental, social, and expressive process whose value is not transferred to someone who simply receives a generated artifact.

“Access to a finished audio file and access to musical craft are two different things, with a substantial body of work in music psychology and education treating musical production as a developmental, social, and expressive process whose value is not transferred to someone who simply receives a generated artifact (Hallam 2010).”

• cites Hallam 2010Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

This could potentially result in both a technical narrowing of the feature space separating AI from human outputs, as well as a cultural narrowing of the genre itself, weakening watermarking strategies that depend on stable separation, and accelerating the genre drift described above.

“This could potentially result in both a technical narrowing of the feature space separating AI from human outputs, as well as a cultural narrowing of the genre itself, weakening watermarking strategies that depend on stable separation, and accelerating the genre drift described above.”

May combine two claims.

Approximately 94% of training material originates from Western musical traditions, while African music accounts for roughly 0.3%, the Middle East 0.4%, and South Asia 0.9%.

“one analysis estimates that approximately 94% of training material originates from Western musical traditions, while African music accounts for roughly 0.3%, the Middle East 0.4%, and South Asia 0.9% (Mehta, Chauhan, and Choudhury 2024).”

• cites Mehta, Chauhan, and Choudhury 2024Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S4 (based).

Human baselines are constrained by 30-second preview availability, which may omit variation present in full tracks.

“Human baselines are constrained by 30-second preview availability, which may omit variation present in full tracks.”

Classified as probably not a substantial claim.

The analysis relies on standard MIR features, which may underrepresent musically salient but hard-to-formalize distinctions.

“Our analysis relies on standard MIR features, which may underrepresent musically salient but hard-to-formalize distinctions (e.g., groove feel, expressive microtiming).”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

This could also shift public expectations toward the platform-legible version of a genre, smoothing away the internal variation that human musicians use to create feel and identity.

“This could also shift public expectations toward the platform-legible version of a genre, smoothing away the internal variation that human musicians use to create feel and identity.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Our study audits two commercially deployed systems across four genres under specific prompting regimes, but does not identify causal mechanisms behind the observed patterns.

“Our study audits two commercially deployed systems across four genres under specific prompting regimes, but does not identify causal mechanisms behind the observed patterns, a common constraint of black-box audits (Metaxa et al. 2021).”

• cites Metaxa et al. 2021Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Over 93% of researchers in the surveyed corpus primarily focus on music from the Global North, while research contributions from Global South institutions remain substantially underrepresented.

“Mehta, Chauhan, and Choudhury (2024) find that over 93% of researchers in their surveyed corpus primarily focus on music from the Global North, while research contributions from Global South institutions remain substantially underrepresented.”

• cites Mehta, Chauhan, and Choudhury 2024Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

The US Copyright Office has reaffirmed that copyright requires human authorship: purely AI-generated material is not copyrightable, although works incorporating AI-generated material may qualify where sufficient human creative authorship is present.

“The US Copyright Office has since reaffirmed that copyright requires human authorship: purely AI-generated material is not copyrightable, although works incorporating AI-generated material may qualify where sufficient human creative authorship is present (US Copyright Office 2025).”

• cites US Copyright Office 2025Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Normalization may not require conscious preference; saturation alone may be sufficient to unconsciously reshape listener expectations.

“so normalization may not require conscious preference; saturation alone may be sufficient to unconsciously reshape listener expectations.”

May combine two claims.

Rights-holders with large catalogs and the legal capacity to negotiate licensing terms are positioned to benefit from compensation arrangements, while independent and emerging artists are not.

“Rights-holders with large catalogs and the legal capacity to negotiate licensing terms are positioned to benefit from that arrangement; independent and emerging artists, who own neither the catalog nor the leverage to extract compensation for its use, are not.”

May combine two claims.

The optimistic counterpoint, that human music becomes more valuable as AI output homogenizes, is structurally plausible but sits against current incentive structures.

“The optimistic counterpoint, that human music becomes more valuable as AI output homogenizes, is structurally plausible but sits against current incentive structures.”

May combine two claims.

Groove in funk drumming, swing in jazz phrasing, and timing deviations in regional hip-hop production are musically salient within practitioner communities yet only partially captured by the spectral and rhythmic features our classifier relies on.

“Groove in funk drumming, swing in jazz phrasing, and timing deviations in regional hip-hop production are musically salient within practitioner communities yet only partially captured by the spectral and rhythmic features our classifier relies on (Keil 1987).”

• cites Keil 1987Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S2 (checkable).

Algorithmic homogenization is a multifaceted justice problem.

“Taken together, the justice frameworks developed above suggest that algorithmic homogenization is not a niche, single-issue concern, but a multifaceted justice problem.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The 2023 Hollywood writers' strike offers a useful precedent in another creative domain; the central issue is not AI use itself, but who controls its terms.

“The 2023 Hollywood writers' strike offers a useful precedent in another creative domain; the central issue is not AI use itself, but who controls its terms (Dwivedi and Chhonkar 2025).”

• cites Dwivedi and Chhonkar 2025Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Their distance from human corpora is detectable and learnable, which complicates simple accounts of transformation while also showing why conventional derivative-work analysis may struggle to capture the kind of genre-level extraction and recombination at issue.

“Their distance from human corpora is detectable and learnable, which complicates simple accounts of transformation while also showing why conventional derivative-work analysis may struggle to capture the kind of genre-level extraction and recombination at issue.”

May combine two claims.

If production speed increases by an order of magnitude, it could alter demand for some forms of production work and affect existing pathways for developing expertise.

“If production speed increases by an order of magnitude, it could alter demand for some forms of production work and affect existing pathways for for developing expertise.”

May combine two claims.

The music continues to circulate, yet the infrastructure no longer depends on the people who produced it.

“The music continues to circulate, yet the infrastructure no longer depends on the people who produced it.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Human reference corpora are constrained to 30-second previews, which may omit variation present in full tracks and could understate within-genre human diversity.

“In our experiments, human reference corpora are constrained to 30-second previews, which may omit variation present in full tracks and could understate within-genre human diversity.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Human producers may intentionally or unintentionally begin adopting the same acoustic signatures prevalent in AI music to optimize for platform visibility and engagement; homogenization may become self-reinforcing.

“A related concern is that human producers may intentionally or unintentionally begin adopting the same acoustic signatures prevalent in AI music to optimize for platform visibility and engagement; that is, homogenization may become self-reinforcing.”