Harmonic Frontier Audio - Non-Lexical Emotional Vocalizations (Preview Pack v0.9)

Harmonic Frontier Audio – Non-Lexical Emotional Vocalizations (Preview, v0.9) A high-fidelity human vocal dataset designed for AI training, speech research, and expressive voice modeling. Non-Lexical Emotional Vocalizations (Preview), created by Harmonic Frontier Audio, provides a compact reference set demonstrating the quality, formatting, and metadata conventions used in the Harmonic Frontier Audio Human Vocality Primitives series. 🔎 Summary This dataset provides high-quality, rights-cleared recordings of non-lexical emotional vocalizations — speech-related affective sounds captured without recognizable words or theatrical exaggeration. The full dataset encompasses: short laughter bursts with irregular airflow pulses sustained laughter with rhythmic airflow and fluctuating pitch soft and heavy sighs with contrasting airflow pressure and release rhythmic sobbing with pulsed airflow and unstable pitch soft whimpers with light phonation and fluctuating pitch contours low-register guttural aversion/disgust bursts sharp surprise gasps with rapid ingressive airflow relief-associated transitions from held breath into extended exhalation These characteristics make the dataset valuable for affective AI synthesis, AI speech and voice modeling, emotion-aware audio generation, phonetics research, non-verbal human sound modeling, embodied-agent behavior, and human-aligned vocal control systems. Developed by Harmonic Frontier Audio, this preview follows The Proteus Standard™ for dataset provenance, transparency, and ethical AI use.Learn more about the Proteus Standard → https://harmonicfrontieraudio.com/proteus-standard Full dataset details and licensing information are available at:https://harmonicfrontieraudio.com/datasets/non-lexical-emotional-vocalizations If you find this dataset useful, please consider giving it a 🤍 on Hugging Face to help others discover it. 🎭 About Non-Lexical Emotional Vocalizations Non-lexical emotional vocalizations are affective human sounds that communicate or accompany emotional states without forming recognizable words. Their acoustic structure can involve rapid airflow pulses, changing pitch, unstable phonation, altered respiratory pressure, abrupt inhalation, or extended exhalation. This dataset spans multiple expressive and physiological behaviors, including laughter, sighing, sobbing, whimpering, guttural aversion gestures, surprise gasps, and relief-linked exhalation. Conditions were produced naturally and speech-like, with emotional intensity reset between takes to preserve controlled variation without drifting into theatrical performance. These phenomena are foundational to: affective and emotion-aware voice synthesis non-verbal human communication modeling expressive conversational-agent behavior multimodal and embodied AI systems acoustic study of emotion-linked airflow and phonation generative modeling of speech-adjacent human vocal behavior This dataset presents a non-linguistic, speech-related representation of emotional vocal behavior.It is not designed as acted dialogue or semantic speech, but rather to isolate acoustic primitives underlying natural affective vocalization and emotion-linked respiratory gestures. 📂 Contents Audio Files (.wav) Recorded at 96 kHz / 24-bit WAV format Exported as mono Fade-ins and fade-outs of 3–5 ms applied for consistency No compression, normalization, or creative processing applied High-pass filtered at ~40 Hz to remove subsonic rumble This preview includes 2 representative audio files, selected to demonstrate: a short laughter burst with irregular airflow pulses and high spectral variability a low-register guttural disgust/aversion burst with expressive non-lexical phonation Metadata (.csv) Includes structured fields for: file name sound source type airflow type phonation type gesture and articulation descriptors microphone and recording chain sample rate, bit depth, and dataset version Metadata follows the Harmonic Frontier Audio – Foundations schema. 🎤 Recording Notes Recorded in a treated studio environment using a single-mic setup: Microphone: Rode NT1-A condenser microphone Recording chain: Rode NT1-A → Zoom F8n Pro Captured at 96 kHz / 32-bit float, rendered as 96 kHz / 24-bit mono WAV for release. Performer positioned approximately 3.5 inches from the microphone, with the microphone approximately 10–15 degrees off-axis. Burst conditions were recorded at approximately 1–2 seconds, with sustained emotional vocalizations approximately 4–8 seconds. Performances were kept natural, speech-related, and non-verbal, with recognizable words avoided. Emotional intensity was reset between takes to preserve authentic variation while avoiding theatrical exaggeration. 🌈 Spectrogram Preview Below is a spectrogram illustrating the irregular pulse structure, transient energy, spectral variability, and expressive phonatory texture characteristic of short laughter and guttural non-lexical emotional vocalizations: ⚡ Usage This preview pack is designed for: Evaluation of Harmonic Frontier Audio dataset quality and structure Testing AI and DSP systems that model affective, emotional, and non-lexical human vocal behavior Research in affective computing, phonetics, expressive voice synthesis, and non-verbal communication Creative and technical applications requiring natural laughter, sighs, sobs, gasps, and other emotion-linked vocal primitives 👉 Note: This is not a full dataset.The complete Non-Lexical Emotional Vocalizations dataset includes short and sustained laughter, soft and heavy sighs, sobbing, whimpering, guttural aversion gestures, surprise gasps, and relief-associated exhalation transitions, and is available for licensing. 💡 Full Dataset Availability This is a preview pack of the Non-Lexical Emotional Vocalizations Dataset.The complete dataset is available for commercial licensing. For licensing inquiries:📩 [e-mail] 📥 How to Use This Dataset in Python You can load the Parquet-converted version of this dataset directly with the datasets library: from datasets import load_dataset dataset = load_dataset( "Harmonic-Frontier-Audio/Non_Lexical_Emotional_Vocalizations_Preview", split="train" ) print(dataset) ⚙️ Note: Parquet conversion and load_dataset() support will be available within 2–3 days of publication. 🔗 Explore More from Harmonic Frontier Audio Human Vocality Primitives Series (Previews) Celtic Constellation Series (Previews) Extended Vocal Techniques Spectrum (Previews) Novelty Gems Cabinet (Previews) (All datasets follow The Proteus Standard™ for ethical dataset provenance and licensing.) 📜 License Released under CC BY-NC 4.0. Free for non-commercial use, testing, and research Commercial licensing available via Harmonic Frontier Audio A formal rights declaration is included in this dataset bundle 📧 Contact Harmonic Frontier Audio📩 [e-mail]🌐 https://harmonicfrontieraudio.com/ 🗒️ Release Notes Version 0.9 (March 2026) – Initial Preview Pack release for Non-Lexical Emotional Vocalizations.See CHANGELOG.md for detailed version history. Citation If you use this dataset in your research, please cite: Pullen, B. (2026). Non-Lexical Emotional Vocalizations Dataset (Preview) [Data set]. Harmonic Frontier Audio. Zenodo. https://doi.org/10.5281/zenodo.21989979 ORCID: https://orcid.org/0009-0003-4527-0178 BibTeX @dataset{pullen_2026_nonlexicalemotionalvocalizations_preview, author = {Blake Pullen}, title = {Non-Lexical Emotional Vocalizations Dataset (Preview)}, year = {2026}, publisher = {Harmonic Frontier Audio}, version = {0.9}, doi = {10.5281/zenodo.21989979}, url = {https://doi.org/10.5281/zenodo.21989979} }