cff-version: 1.2.0
message: "If you use this dataset, please cite it as below."
type: dataset
title: "TonalityPrint: A Contrast-Structured Voice Dataset for Exploring Functional Tonal Intent, Ambivalence, and Inference-Time Prosodic Alignment v1.0"
version: 1.0.0
doi: 10.5281/zenodo.17913895
date-released: 2026-01-24
url: "https://doi.org/10.5281/zenodo.17913895"
repository-code: "https://github.com/TonalityPrint/TonalityPrint-v1"
license: CC-BY-NC-4.0
authors:
  - family-names: Polhill
    given-names: Ronda
    email: ronda@TonalityPrint.com
    affiliation: Independent Researcher
    orcid: ""
keywords:
  - tonality
  - inference
  - ambivalence detection
  - functional tonal intent
  - voice dataset
  - prosody dataset
  - human-AI communication
  - conversational AI
  - single speaker dataset
  - tonality regression
  - voice AI
  - voice alignment
  - fine tuning
  - AI safety
  - autonomous systems
  - sycophancy-mitigation
  - voice agents
  - personalized AI
  - embodied AI
  - companion AI
  - ethical voice data
  - expressive synthesis
  - humanoid robotics
  - prosodic interpretability
  - intent aligned dataset
  - inference-time prosodic alignment
  - trust calibration
  - fine-tuning dataset
  - human voice dataset
  - intent drift
  - tonal alignment
  - agentic AI
  - outcome inference
  - human-AI alignment
  - uncanny-valley-effect
  - prosodic trust
  - prosodic intentionality
  - safety alignment
  - prosodic style transfer
  - empathetic AI
  - humanoid voice appearance
  - human-in-the-loop
  - human baseline
  - real-world experience
abstract: >
  TonalityPrint is a specialized single-speaker speech corpus designed
  to enable 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. The dataset provides 144 audio samples across 18
  utterances, each recorded in 8 parallel prosodic states. A core innovation
  is systematic ambivalence annotation, treating tonal complexity as a
  perceptual entropy cross-intent feature rather than noise. Utilizing a
  Fixed-Phrase Octet design, the dataset enables Differential Latent Analysis
  (DLA) for contrastive approximation of tonal intent vectors. Annotations
  are grounded in 8,873+ consequential interactions, capturing an AI-adjacent
  yet trusted vocal profile. TonalityPrint is intended as a hypothesized
  contrast substrate for researchers exploring inference-time alignment,
  prosodic interpretability, style-conditioned synthesis, human-AI voice
  calibration, and safety-critical voice agents. All recordings are 100%
  authentic human voice (author) with explicit consent, released under CC
  BY-NC 4.0 (academic/research free; commercial licensing available).
references:
  - type: article
    title: "Tonality as Attention"
    authors:
      - family-names: Polhill
        given-names: Ronda
    year: 2025
    publisher:
      name: Zenodo
    doi: 10.5281/zenodo.17410581
contact:
  - email: ronda@TonalityPrint.com
    name: Ronda Polhill