Problem
Research questions and friction points this paper is trying to address.
music generation
attribution
memorization detection
audio source verification
copyright
Tracing Inputs, Verifying Outputs: Validating Attribution in Music Generation
🤖 AI Summary
This study addresses the challenges of verifying contributor attribution and mitigating training data memorization leakage in AI music generation by proposing a complementary attribution framework conditioned solely on audio. Methodologically, the approach integrates a MixAudio generator, a musicDNA version identification model, and controlled input swapping techniques to enable precise tracing of input-to-output influence and deduplication auditing without textual prompts. Experimental results demonstrate that the proposed model faithfully adheres to input conditions across timbral and harmonic dimensions, achieving significantly higher human-evaluated precision and recall compared to existing detectors. This research provides verifiable technical evidence to support copyright compensation mechanisms for AI-generated music.
📝 Abstract
How can we verify whose music contributed to an AI-generated output? This paper demonstrates how input-based attribution can provide verifiable evidence of which audio sources were used in a generation and whether they shaped the output. To do so, we condition the generation solely on audio without any text input, then trace the inputs behind each output, and establish their musical effect. In prompt adherence tests and controlled input swaps, the stems generated by our generator, MixAudio, follow the prompt audio in timbre and the context audio in harmony. Yet these outputs may still reproduce training data not supplied as inputs. We therefore audit memorization with our musical version identification model, musicDNA, and find few reproductions outside the input records. On human-judged cases within the flagged pool, it achieves higher precision and recall than the other tested memorization detectors. The two evaluations suggest that input records and output analysis provide complementary evidence for attribution, on which rights-holder reporting and compensation can draw as the AI music economy takes shape. Audio examples are available at https://neutune.github.io/attr2027demo/
Innovation
Methods, ideas, or system contributions that make the work stand out.
Music Generation
Attribution Verification
Memorization Detection
Audio Conditioning
Music Version Identification
🔎 Similar Papers
No similar papers found.
