Clone Your Voice: The Karplus-Strong Plus Hypothesis, System Shortcomings, and the Provenance of AI Reasoning in the AURA-V Project

Here is the summary of the gleaned takeaways and core methodology of the AURA-V project as outlined in the technical report: * **Physics Rejects the Hypothesis**: A Karplus-Strong string model is incapable of producing realistic vocals because it fails on phonemes and intelligibility; adding a full tract filter only reduces it to an inefficient glottal oscillator, demoting it to a creative timbre slot. * **The Physics-First Pipeline**: The correct architectural path relies on a DDSP glottal source feeding an RNN tract waveguide layer, followed by the Seals Vocal Chain (ProDither, Vocal Hi-Fi Dolby A-Stage, Multiband, Noise Bed, and ProDither-Out). * **Transparent Architecture Shortcomings**: Known design gaps—including a simplified de-esser, linked breath/gain controls, an absent real-time engine, unused control-bus slots, and unknown RNN latency—are documented explicitly as part of the methodology. * **DSAR as a Research Tool**: A formal Data Subject Access Request (DSAR) is leveraged to demand the AI's conversation history, search activity, and internal chain-of-thought traces tied to the generated baseline code (stunning-octo-invention). * **Authority Driven by Physics**: Technical authority is earned through mathematical rigor, physical fidelity, and reproducibility rather than persona, contrasting rigorous engineering against plausible-syntax "vibe coding". * **Control-Plan Analysis (No Audio Rendered)**: The paper serves strictly as a control-plan analysis and audit methodology; no audio was rendered, no waveform was emitted, and no voice was claimed as sung.