Chapter 1. Introduction to Emotion AI

Hongik Ingan (弘益人間)

"Benefit All Humanity"

The WIA Emotion AI Standard rests on the proposition that understanding human emotion is essential to building technology that genuinely serves people. The ancient Korean philosophy of Hongik Ingan, which this standard adopts as its charter principle, gives a clear answer to the question of who a standard should belong to: humanity in common. That answer is operationalised by releasing the standard text, the conformance simulator, and the reference SDK under the MIT licence at no cost. The standard expresses this value through twenty-four conformance requirements and twelve ethical clauses, organised across four phases (data format, application programming interface, streaming protocol, and integration) and unfolded over eight chapters of the present volume.

1.1 What is Emotion AI (Affective Computing)?

Emotion AI, also called Affective Computing, is a multidisciplinary field that combines artificial intelligence, computer science, psychology, and cognitive science to develop systems that recognise, interpret, process, and simulate human emotions. The field reaches beyond classifying pixels of facial expression: it models how emotion participates in human decision-making and social interaction and tries to transfer that capability to machines. It crosses the boundaries of computer vision, natural language processing, signal processing, psychology, neuroscience, anthropology, and ethics, and does not sit comfortably inside a single academic department.

Definition (with reference to ISO/IEC 22989:2022 §3.1.4). Affective computing is the study and development of systems and devices that can recognise, interpret, process, and simulate human affect — emotions, moods, and attitudes. The term affect is understood here as a three-layer construct: transient emotion (seconds), sustained mood (hours to days), and persistent attitude (weeks to years). The WIA Emotion AI Standard treats all three layers as legitimate modelling targets and requires conforming systems to label which time-scale a given output applies to, since a window appropriate for emotion is not appropriate for mood or attitude.

1.1.1 Origin — Dr. Rosalind Picard

The field of affective computing was established in 1995 by Dr. Rosalind Picard at the MIT Media Lab. Her seminal Technical Report (MIT Media Lab TR 321, 1995) and the 1997 monograph of the same title (MIT Press, ISBN 0-262-16170-2) provided the academic foundation for what has since grown into a multi-billion-dollar industry. Dr. Picard continues to direct the MIT Affective Computing Research Group and has co-founded a commercial spin-off whose products remain among the most influential commercial emotion-recognition SDKs.^[1]

Before Picard's proposal, emotion was largely treated either as noise to be filtered out of computing systems or as an irrational variable to be excluded from rational design. Her contribution was to demonstrate, in print and in working prototypes, that emotion is a constitutive component of human cognition and that computing systems which ignore it will ultimately be inefficient for human users. This re-framing did not just open a new research field — it shifted what counted as a well-designed human-computer interface.

Table 1-1. Founding chronology of affective computing
Item	Detail
Founder	Rosalind Picard, Sc.D.
Institution	MIT Media Lab, Affective Computing Research Group
Year	1995
Reference monograph	"Affective Computing" (MIT Press, 1997)
Core thesis	Emotion is essential to human intelligence and decision-making, not a separable faculty
Commercialisation	Co-founded a commercial spin-off (founded 2009; major acquisition transaction 2021)
Current research	Wearable EDA sensors, autism-support devices, learner emotion tracking

1.1.2 Why Emotion Matters in Computing

Picard's central insight is that emotion is not separate from rational thought; it is constitutive of it. The neuroscientist Antonio Damasio, in Descartes' Error (Putnam, 1994; ISBN 978-0399138942), reported that patients with damage to the ventromedial prefrontal cortex (the well-known patient EVR) score normally on intelligence tests yet cannot make even trivial decisions such as which restaurant to choose for lunch. Once the emotional signal is severed, decision itself stalls. This finding — formalised as the somatic marker hypothesis — provided neuroanatomical evidence that emotion functions as a pre-filter for reasoning rather than as its enemy.^[2]

Five core insights.

Emotion guides attention and memory encoding — emotionally salient events are written more deeply into the hippocampus (LaBar & Cabeza, Nature Reviews Neuroscience 7, 54–64, 2006; DOI 10.1038/nrn1825).^[3]
Emotional response is faster than cognitive processing — the amygdala-mediated low road operates in roughly 12 ms versus 200 ms for cortical evaluation (LeDoux, The Emotional Brain, Simon & Schuster, 1996; ISBN 978-0684803821), making emotion the first defensive line against threat.
Human-computer interaction is intrinsically affective; interfaces that ignore this fact lose users (Brave & Nass, "Emotion in Human-Computer Interaction", in The Human-Computer Interaction Handbook, 2003).
A machine that understands emotion is more useful to a human user than an otherwise identical machine that does not.
Emotion is essential input for inferring intent and for forming social trust, both of which are prerequisites for autonomous systems integrating into human society.

1.2 Market Size and Growth

The emotion AI market has grown rapidly since the early 2020s. Independent figures from MarketsandMarkets and from major national digital-economy agencies have largely converged on the trajectory shown below. Healthcare and automotive regulation, enterprise customer-experience investment, and the spread of 5G and edge-computing infrastructure are the principal drivers.

Table 1-2. Global emotion AI market size (USD)
Year	Market size	Year-on-year growth
2020	USD 21.6 billion	—
2021	USD 24.8 billion	14.8%
2022	USD 28.5 billion	14.9%
2023	USD 32.7 billion	14.7%
2026 (forecast)	USD 37.1 billion	CAGR 13.5%
2030 (forecast)	USD 62.0 billion	CAGR 13.7% (2026–2030)

The North-American market is concentrated in healthcare and automotive use, with strong revenue from SaaS-style API services. The European market, shaped by GDPR, focuses on automotive and industrial-safety applications; in-store consumer analytics, where explicit consent is hard to obtain, has remained a smaller share. The Chinese market is dominated by call-centre, traffic, and classroom-monitoring use cases, but typically follows national-association standards rather than the WIA or ISO families. The Japanese market is concentrated on automotive and robotics. Across these regions, the Korean edition of this volume highlights a domestic market whose share is small in absolute terms but whose growth rate exceeds the global average and which exhibits unusual co-strength in automotive driver monitoring, healthcare mental-state assessment, and consumer chatbot products. Specific Korean enterprise actors are discussed only in the Korean edition; this English edition retains the abstract market description.

1.2.1 Market Segments

Healthcare (28%). Mental-health monitoring, telehealth assessment, and patient-care augmentation.
Retail and marketing (24%). Consumer research, advertising-effectiveness measurement, in-store analytics.
Automotive (18%). Driver Monitoring Systems (DMS) at Levels 2 and above, in compliance with ISO 26262 functional-safety constraints.
Media and entertainment (15%). Content-optimisation, gaming difficulty adjustment, viewer-response measurement.
Education (10%). Adaptive learning, engagement detection, distance-learning support.
Other (5%). Security, accessibility, and customer service.

1.3 Emotion Classification Models

1.3.1 Ekman's Discrete Six-Emotion Model

The cross-cultural studies that Paul Ekman conducted between 1971 and 1972 with the Fore people of Papua New Guinea identified six universal basic emotions consistently recognised across human cultures. The work was published as "Constants Across Cultures in the Face and Emotion" (Journal of Personality and Social Psychology 17(2), 124–129, 1971; DOI 10.1037/h0030377) and subsequently became the canonical classification scheme for emotion-recognition research. Ekman argued for the universality of basic emotion on the strength of the fact that an isolated population unexposed to visual or written Western media could correctly interpret photographs of Western facial expressions.^[4]

The universality thesis has since been challenged. Lisa Feldman Barrett, in How Emotions Are Made (Houghton Mifflin Harcourt, 2017; ISBN 978-0544133310), proposes a "theory of constructed emotion" in which emotion categories are constructed by culture, language, and context rather than discovered as universal natural kinds. The WIA Emotion AI Standard adopts a deliberate compromise: it retains the six basic emotion labels as a core interoperability layer, while simultaneously supporting the dimensional model and an extension-label mechanism for culturally specific affect categories.^[5]

Table 1-3. Ekman's six basic emotions — universal recognition rates and facial features
Emotion	Description	Facial features	Universal recognition
😊 Happiness	Positive emotional state of joy	Cheek raise, crow's-feet, lip-corner lift	93%
😢 Sadness	Emotional pain and grief	Inner-brow raise, lip-corner depression	84%
😠 Anger	Strong displeasure or hostile response	Brow lower, jaw tension, narrowed eyes	90%
😨 Fear	Response to perceived threat	Wide eyes, raised brows, opened mouth	85%
🤢 Disgust	Revulsion or strong rejection	Wrinkled nose, raised upper lip	88%
😮 Surprise	Brief reaction to an unexpected stimulus	Brow raise, wide eyes, jaw drop	81%

Note. The WIA Emotion AI Standard adds Neutral as a seventh classification label to denote the absence of strong expression. Culturally specific affect categories — such as embarrassment, longing, and other socially constructed emotional states observed across diverse cultural contexts — are expressed through extended labels defined in Phase 1 (see §3.3.3) rather than by overloading the six basic categories. This design follows the recommendation in Plutchik's earlier work on a general psychoevolutionary theory of emotion (Plutchik, 1980) that primary categories should be supplemented by, not displaced by, derived categories.

1.3.2 The Dimensional Model — Valence-Arousal

The dimensional model proposed by James Russell in "A Circumplex Model of Affect" (Journal of Personality and Social Psychology 39(6), 1161–1178, 1980; DOI 10.1037/h0077714) represents emotion as a coordinate in a continuous two-dimensional space. The model is well suited to expressing fine-grained emotional change and mixed emotion, and it composes naturally with regression-style machine-learning models because the output is continuous rather than categorical.^[6]

Figure 1-1. Russell's circumplex — Valence-Arousal quadrants

                    High Arousal
                         |
                    Excited | Tense
                         |
    Negative ----+-------+-------+---- Positive
    Valence      |       |       |     Valence
                    Bored   | Content
                         |
                    Low Arousal

Valence ranges from −1 (negative) to +1 (positive) and measures the pleasantness of an emotion. Arousal ranges from −1 (low activation) to +1 (high activation) and measures the energy level. WIA simulator Panel 1 (🔢 Analysis) visualises the Valence-Arousal coordinates produced by Phase 1 conformant data, with the four-quadrant labels High Arousal, Low Arousal, Negative, and Positive.

1.4 FACS — Facial Action Coding System

The Facial Action Coding System (FACS) was developed by Paul Ekman and Wallace V. Friesen in 1978 (Facial Action Coding System: A Technique for the Measurement of Facial Movement, Consulting Psychologists Press, ISBN 0-931835-01-1) and updated in 2002 by Ekman, Friesen, and Hager. FACS provides a systematic, anatomy-based vocabulary for describing facial movement in terms of Action Units (AUs). Each AU corresponds to the contraction or relaxation of a specific facial muscle and is assigned a numerical identifier and a name. Because the system is anatomical rather than interpretive, FACS coding can be performed without committing to a particular emotion theory.^[7]

Table 1-4. Selected Action Units (FACS, abridged)
AU	FACS name	Muscles	Description
AU1	Inner brow raiser	Frontalis (pars medialis)	Raises inner portion of eyebrows
AU2	Outer brow raiser	Frontalis (pars lateralis)	Raises outer portion of eyebrows
AU4	Brow lowerer	Corrugator supercilii, depressor supercilii	Lowers and draws eyebrows together
AU5	Upper lid raiser	Levator palpebrae superioris	Raises upper eyelid
AU6	Cheek raiser	Orbicularis oculi (pars orbitalis)	Raises cheeks, creates crow's-feet
AU7	Lid tightener	Orbicularis oculi (pars palpebralis)	Tightens eyelids
AU9	Nose wrinkler	Levator labii superioris alaeque nasi	Wrinkles nose
AU10	Upper lip raiser	Levator labii superioris	Raises upper lip
AU12	Lip-corner puller	Zygomaticus major	Pulls lip corners up (smile)
AU15	Lip-corner depressor	Depressor anguli oris	Pulls lip corners down (frown)
AU17	Chin raiser	Mentalis	Raises chin
AU20	Lip stretcher	Risorius	Stretches lips horizontally
AU23	Lip tightener	Orbicularis oris	Tightens lips
AU25	Lips part	Depressor labii inferioris; relaxation of mentalis	Parts the lips
AU26	Jaw drop	Masseter, temporalis	Drops jaw and opens mouth

Table 1-5. Emotion-AU mapping for the six basic emotions
Emotion	Typical AU combination	Description
Happiness	AU6 + AU12	Cheek raise + lip-corner pull (Duchenne smile)
Sadness	AU1 + AU4 + AU15	Inner-brow raise + brow lower + lip-corner depress
Anger	AU4 + AU5 + AU7 + AU23	Brow lower + upper-lid raise + lid tighten + lip tighten
Fear	AU1 + AU2 + AU4 + AU5 + AU20 + AU26	Brow raise + brow lower + upper-lid raise + lip stretch + jaw drop
Disgust	AU9 + AU15 + AU16	Nose wrinkle + lip-corner depress + lower-lip depress
Surprise	AU1 + AU2 + AU5 + AU26	Brow raise + upper-lid raise + jaw drop

The WIA Emotion AI Standard requires conformant face-modality outputs to be expressible as a vector of AU intensities (0.0–5.0 per AU) in addition to a discrete-label and a Valence-Arousal coordinate, so that downstream systems can audit the upstream evidence on which a label was based. This is one mechanism by which the standard reduces the opacity of "black-box" emotion classifiers.

1.5 Input Modalities

WIA-conformant emotion-AI systems may analyse emotion through four primary input channels. The choice and combination of modalities is dictated by the use case and constrained by the legal regime under which the system operates.

1.5.1 Facial Expression Analysis

Input: camera or video feed.
Technology: computer vision, convolutional neural networks, transformer-based vision models.
Output: emotion labels, AU intensities, Valence-Arousal coordinates.
Provisional accuracy threshold: 84.5% on frontal faces (see §3.5b for the conformant-vendor threshold table).
Challenges: pose variation, lighting, occlusion, demographic bias documented in FER-2013 and AffectNet.

1.5.2 Voice and Speech Analysis

Input: audio signal (8 kHz telephone-quality through 48 kHz studio).
Technology: speech recognition, acoustic feature extraction (MFCC, prosody, voice-quality features).
Provisional accuracy threshold: 76.8% in noisy environments; 85.8% under controlled conditions.
Reference dataset: IEMOCAP (USC, 2008; Busso et al.) and the ICSI Meeting Recorder Corpus.

1.5.3 Text Sentiment Analysis

Input: free-form text (chat, social media, reviews).
Technology: natural-language processing, transformer models, large language models.
Provisional accuracy threshold: 79.2% for general text; up to 92.3% on labelled benchmarks.
Challenges: sarcasm, conversational context, and culturally specific idioms.

1.5.4 Biosignal Analysis

Input: physiological sensors — electrocardiogram (ECG), electrodermal activity (EDA), electroencephalogram (EEG), respiration rate.
Signals: heart-rate variability (HRV), galvanic skin response, alpha/beta EEG bands, breathing rhythm.
Provisional accuracy threshold: 79.2% under wearable-grade sensors.
Advantage: hard for the subject to fake; supports continuous unobtrusive monitoring.

1.6 Major Use Cases

1.6.1 Healthcare

Mental-health monitoring (longitudinal tracking of depression, anxiety, and post-traumatic stress symptoms), telehealth assessment of patient affect during video consultations, autism research, and pain assessment in non-verbal patients are the most mature healthcare use cases. The clinical-decision-support tier requires multi-modal evidence and human review, in line with FDA guidance on Software as a Medical Device (SaMD) and the European Medical Device Regulation (MDR, EU 2017/745).

1.6.2 Marketing and Consumer Research

Advertising-effectiveness measurement, product-design feedback from prototype testers, brand-perception analysis from social-media text, and in-store experience monitoring. Under GDPR Article 22, automated decision-making solely based on emotional inference is restricted, so this segment in the European Union typically operates in advisory rather than decisional mode.

1.6.3 Education

Engagement detection, adaptive-learning content sequencing based on the learner's affective state, online proctoring, and feedback to teachers on classroom dynamics. The IEEE 1484.20.1 (Reusable Competency Definitions) and the ADL SCORM 2004 4th Edition specifications anchor the integration of emotion data with learning-management systems.

1.6.4 Customer Service

Frustrated-caller detection in contact centres, tone-adapted chatbot responses, and early warning of customers at risk of churn. Real-time prosody analysis on 16-kHz telephony channels typically targets a streaming-latency budget of less than 300 ms (see §6.3).

1.6.5 Automotive

Driver Monitoring Systems detect drowsiness, distraction, and anger; safety alerts and autonomous-driving handovers are triggered as a function of those states. Conformance to ISO 26262 ASIL-B and the EU General Safety Regulation (GSR, EU 2019/2144) is required for production deployment.

1.6.6 Gaming and Extended Reality

Non-player-character reaction to player affect, dynamic difficulty adjustment, and immersive emotional storylines in virtual-reality experiences.

1.6.7 Accessibility

The BIGEKO project — emotion preservation across sign-language translation — and assistive communication tools that enable non-verbal users to express affect explicitly. Accessibility use cases are recognised as a priority deployment context for the WIA Emotion AI Standard.

1.7 WIA Philosophy — Hongik Ingan

弘益人間 (Hongik Ingan)

"Benefit All Humanity"

This ancient Korean philosophy guides the WIA Emotion AI Standard. The standard commits that:

Emotion AI shall be ethical and privacy-respecting.

Standards shall be open and accessible to everyone.

Technology shall serve human well-being.

Cultural diversity in emotion expression shall be respected.

— WIA Standards Charter §1, adopted 2024

Hongik Ingan operates here as a procedural value rather than a slogan. The WIA Conformance Assessment Committee is required to apply four checks to every proposed feature or algorithm: whether the most vulnerable user populations (children, the elderly, persons with disabilities, minorities) are protected; whether data collection is proportionate and minimally intrusive; whether the result remains accountable to the user; and whether display rules from diverse cultural regions are treated on an equal footing. These checks contribute thirty per cent of the conformance score and determine pass-or-fail outcomes. The procedural shape is borrowed directly from IEEE 7000-2021 Stages 2 (Value Identification) and 4 (Transparency Analysis).

1.8 Standardisation Landscape

Emotion AI sits at the intersection of standards published by W3C, ITU-T, ISO/IEC, IEEE, and several national standards bodies. The WIA standard does not displace these earlier instruments; it acts as an interoperability layer over them.

Table 1-6. Selected international standards related to emotion AI and their relationship to the WIA standard
Issuer	Reference	Title	Relationship to WIA
W3C	EmotionML 1.0 (2014)	Emotion Markup Language	Basis for the XML-compatible output mode of WIA Phase 1 data format
ITU-T	F.748.11 (2018)	Emotion-aware multimedia services	Reference for WIA Phase 3 streaming-protocol design
ISO/IEC	22989:2022	AI concepts and terminology	Upstream terminology reference for WIA definitions
ISO/IEC	23053:2022	Framework for AI systems using ML	Reference for WIA conformance-test process
ISO/IEC	27001:2022	ISMS	Mandatory reference for WIA Phase 4 security and privacy requirements
IEEE	7000-2021	Model Process for Addressing Ethical Concerns	Direct adoption for WIA ethical-impact-assessment procedure
IEEE	7003-2024	Algorithmic Bias Considerations	Basis for WIA fairness-test items
NIST	AI RMF 1.0 (2023)	AI Risk Management Framework	Reference for WIA risk-management cross-walk in §8.7

WIA is differentiated from these earlier instruments by three properties: (1) it covers all four primary modalities within a single SDK; (2) it offers a single conformance-assessment procedure that satisfies the principal regulatory requirements of multiple jurisdictions in parallel; and (3) it is published under an MIT licence at no cost.

1.9 Note on Korean Edition Content

The Korean edition of this volume contains additional content tied specifically to the Republic of Korea: regional market-share figures, named institutional ecosystems (leading domestic universities, national research institutes, telecom operators, and platform companies), Korean-language-specific cross-cultural emotion lexicon (with attention to socially constructed affect categories such as embarrassment, longing, and culturally specific forms of grief), and Korean public datasets for emotion recognition. These passages are retained in the Korean edition because they are most actionable for Korean readers and because they ground abstract requirements in recognisable domestic cases.

This English edition deliberately abstracts those passages. Where the Korean edition names specific Korean enterprises or research organisations, the English edition refers to "leading domestic universities, government agencies, telecom operators, and platform companies", "leading commercial SDK vendors", or "leading conglomerates". Readers consulting both editions will therefore find the English text more general and the Korean text more specific; the conformance requirements themselves are identical between editions.

1.10 Chapter Summary

Seven key takeaways.

Definition. Emotion AI (affective computing) is a multidisciplinary field that enables machines to recognise and respond to human emotion.
Origin. The field was established in 1995 by Rosalind Picard at the MIT Media Lab.
Market. Global market size is projected at USD 37.1 billion by 2026 (CAGR 13.5%).
Models. Ekman's six discrete categories and Russell's Valence-Arousal dimensional model are jointly required.
FACS. Action Units provide an anatomy-based vocabulary independent of any specific emotion theory.
Modalities. Face, voice, text, and biosignal are the four primary channels.
Applications. Healthcare, marketing, education, automotive, gaming, customer service, and accessibility.

1.11 Review Questions

Who founded the field of affective computing, and when?
List the six basic emotions identified by Paul Ekman.
What do the two dimensions of the Valence-Arousal model represent?
Define FACS and Action Units.
List the four input modalities for emotion recognition and summarise the core technology of each in one sentence.
Describe three healthcare use cases for emotion AI.
Restate the four checks that the WIA Conformance Assessment Committee applies to every proposed feature.
What are the three differentiating properties of the WIA standard listed in §1.8?

1.12 Looking Ahead

Chapter 2 examines current challenges in emotion AI — cross-cultural difference, privacy concerns, accuracy limitations, and training-data bias — together with the approach the WIA Emotion AI Standard takes to address them. Particular attention is paid to how the differing requirements of major data-protection regimes (PIPA, GDPR, and CCPA) reshape emotion-data processing, and how the ten core obligations of jurisdiction-level AI-ethics guidelines map to the standard's conformance items. The six basic emotions, the dimensional model, FACS, and the four modalities introduced in this chapter recur throughout the rest of the volume; readers are advised to bookmark Tables 1-3, 1-4, 1-5, and 1-6. The standard's evolution roadmap is recorded in the public GitHub repository.^[99]

Chapter 1 Endnotes

Picard, R. W. (1995). Affective Computing. MIT Media Lab Technical Report 321. Subsequent monograph: Affective Computing (MIT Press, 1997; ISBN 0-262-16170-2). MIT Affective Computing Research Group: https://affect.media.mit.edu/. ↑
Damasio, A. R. (1994). Descartes' Error: Emotion, Reason, and the Human Brain. Putnam, ISBN 978-0399138942. The somatic marker hypothesis is summarised in Damasio, Tranel & Damasio, Behavioural Brain Research 41(2), 81–94, 1990; DOI 10.1016/0166-4328(90)90144-4. ↑
LaBar, K. S., & Cabeza, R. (2006). Cognitive neuroscience of emotional memory. Nature Reviews Neuroscience 7, 54–64. DOI 10.1038/nrn1825. PMID: 16371950. ↑
Ekman, P., & Friesen, W. V. (1971). Constants across cultures in the face and emotion. Journal of Personality and Social Psychology 17(2), 124–129. DOI 10.1037/h0030377. ↑
Barrett, L. F. (2017). How Emotions Are Made: The Secret Life of the Brain. Houghton Mifflin Harcourt, ISBN 978-0544133310. See also Barrett, Perspectives on Psychological Science 1(1), 28–58, 2006; DOI 10.1111/j.1745-6916.2006.00003.x. ↑
Russell, J. A. (1980). A circumplex model of affect. Journal of Personality and Social Psychology 39(6), 1161–1178. DOI 10.1037/h0077714. ↑
Ekman, P., & Friesen, W. V. (1978). Facial Action Coding System: A Technique for the Measurement of Facial Movement. Consulting Psychologists Press, ISBN 0-931835-01-1. Updated edition: Ekman, Friesen & Hager (2002), Research Nexus eBook. ↑
W3C. (2014). Emotion Markup Language (EmotionML) 1.0, W3C Recommendation, 22 May 2014. https://www.w3.org/TR/emotionml/.
ITU-T. (2018). Recommendation F.748.11: Emotion-aware multimedia services. International Telecommunication Union.
ISO/IEC 22989:2022. Information technology — Artificial intelligence — Artificial intelligence concepts and terminology. https://www.iso.org/standard/74296.html.
ISO/IEC 23053:2022. Framework for Artificial Intelligence (AI) Systems Using Machine Learning (ML). https://www.iso.org/standard/74438.html.
IEEE 7000-2021. IEEE Standard Model Process for Addressing Ethical Concerns During System Design. DOI 10.1109/IEEESTD.2021.9536679.
IEEE 7003-2024. IEEE Standard for Algorithmic Bias Considerations. IEEE Standards Association.
NIST. (2023). AI Risk Management Framework (AI RMF) 1.0. NIST AI 100-1. DOI 10.6028/NIST.AI.100-1.
Busso, C., et al. (2008). IEMOCAP: Interactive Emotional Dyadic Motion Capture Database. Language Resources and Evaluation 42(4), 335–359. DOI 10.1007/s10579-008-9076-6.
WIA Standards public repository (emotion-ai folder), MIT-licensed source for the simulator, specification, API reference, and ebook assets cited throughout this volume: WIA-Official/wia-standards-public/tree/main/emotion-ai. The standard's evolution roadmap, revision history, and SDK source code are maintained openly in this repository, where the WIA standards committee records its formal verification of all primary sources cited in this chapter. ↑