TonalityPrint: A Contrast-Structured Voice Dataset for Exploring Functional Tonal Intent, Ambivalence, and Inference-Time Prosodic Alignment v1.0

TonalityPrint is a specialized single-speaker speech corpus designed to enable the exploration of fine-tuning functional tonal intents - Trust, Attention, Reciprocity, Empathy Resonance, and Cognitive Energy - in voice AI systems. Unlike emotion recognition datasets, TonalityPrint annotates functional tonal intents (what speakers do with tone), not just what they feel. Annotations include five Functional Tonal Intents and an explicit ambivalence condition, conceptualized as a perceptual entropy transitional state rather than a discrete emotion. A core innovation of TonalityPrint is its treatment of Ambivalence (systematically annotated as ambivalex), where, rather than discarding mixed or transitional signals as noise, this dataset treats tonal complexity as a perceptual entropy feature essential for real-world inference-time alignment. Utilizing its Fixed-Phrase Octet, the dataset delivers 144 audio samples across 18 utterances, each recorded in 8 parallel prosodic states. It is accompanied by a detailed README describing design philosophy, ethical constraints, and proposed evaluation affordances. Grounded in real-world practitioner experience from 8,873+ consequential interactions, the corpus potentially captures an “AI-adjacent yet trusted” vocal profile observational motivation that may challenge assumptions about the ‘uncanny valley’ effects and potentially offer provocative insights for humanoid robotics, companion AI, human-agent interaction and reasoning-based voice interfaces. TonalityPrint is intended as a hypothesized contrast substrate, not a training corpus for general-purpose speech models. TonalityPrint is designed for researchers exploring inference-time alignment, prosodic interpretability, style-conditioned synthesis, human-AI voice calibration, and evaluation of "Safety-Critical" voice agents (e.g., healthcare, autonomous systems) that must audibly sound uncertain when hallucinating. Featuring; •144 high-fidelity, unprocessed WAVs preserve tonal fidelity (48kHz/32-bit). •18 unique utterances across 8 prosodic states. •Continuous intensity indices (0-100) for five core functional intents. •Comprehensive metadata including "ambivalex" flags and practitioner-verified outcome associations. All recordings: 100% authentic human voice (author) with explicit consent; Released under CC BY-NC 4.0 (academic/research free; commercial licensing available). This work emerges from independent practitioner-research conducted without institutional funding and is released for academic research use under CC BY-NC 4.0. Commercial licensing is available. Supplement to: Polhill, R. (2025) "Tonality as Attention" white paper (DOI: 10.5281/zenodo.17410581). Why Download Now: TonalityPrint is designed to enable precision isolation of functional prosodic signals in voice AI - a growing priority for labs focused on safe, nuanced, and human-aligned speech interfaces. Dataset v1.0 is available today for benchmarking; collaborative validation and multi-speaker extensions are actively sought.