Emotional Speech Overview Who cares? The Idea of Emotion Difficulties in approaching Describing Emotion Computational Models Modeling Emotion in Speech An example – Ang ’02 Who Cares? Practical impact Detecting Frustration/Anger Stress/Distress Help call prioritizing Tutorials – Boredom/Confusion/Frustration Pacing/Positive feedback User acceptance Users preferred talking head using ES (Stallo, in Schröder) Who Cares? Esoteric Impact Is artificial intelligence possible w/o detection of emotion? w/o display of “emotion”? Do we experience someone/something as understanding us if it can’t understand our emotional state/experience? Who Cares? – Izard ’77 Emotion & Perception E & Cognition E & Action E & Personality Development Understanding a speaker’s emotional state gives us insight into his/her intention, desire, motivation (Zimring) The Bad News (Picard ’97) Maintaining realistic expectations User’s confidence in information Potential to forge affective channels Problem solving vs. empathic/observational Symmetry of communication Privacy issues Idea of Emotion (Hergenhahn ’01) Descartes “Passions” Understood emotions as originating from both physiological and cognitive sources Pineal gland Late 1800’s – early 1900’s Psychology was study of consciousness William James “The Science of Mental Life” Major method was introspection – mental – Relies on a person reporting his/her experience Idea of Emotion 1930’s – 1950’s Behaviorist tradition – study of behavior “Objective” (at least measurable and observable) Emerged from academia – a lot of rats suffered Explains everything in terms of stimulus / response Fails to explain some crucial issues, e.g., language Idea of emotion No one expects to get wet in a pool filled with ping pong ball models of water molecules. 1950’s – Cognitive “Revolution” Piaget, Miller, Chomsky, et al. Miller The Science of Mental Life John Searle Syntax vs. semantics Materialism vs. Dualism What are reasonable expectations? Searle ’90 Difficulties in approaching (Cowie) E is resistant to capture in symbols Speech presents special problems Modeling of primary E’s not so useful Consensus Display Rules (Ekman) Mixes “Love/hate relationship” Negative response to simulated displays “[Utterances were] said by two actors in the emotions of happiness, sadness, anger […]” Difficulties in approaching Quality of reference data Rating believability (Schröder) Forced choice tests often ignore issue of appropriateness/believability “How appropriate was utterance to given E” (Rank 98) (Iida, et al.) Rated using scales for preference and for subjective degree of expressed E. Subject generosity Temporal and contextual relationships Pereira Everything it is possible to analyze depends on a clear method of distinguishing the similar from the dissimilar. Describing Emotion = = = ≠ – Carl Linnaeus ≠ Invariants Describing Emotion (Cowie) Primary emotions Acceptance, anger, anticipation, disgust, joy, fear, sadness, surprise Secondary Emotions Arousal Attitude An aside: Intention may generate all of these activity decisiveness haughtiness restrained adoration delighted helplessness restraint alarm dependence hope righteousness alertness depression humiliation rigor anger desire indifference routine animosity despair inferiority sadness annoyance dimness initiative satisfaction anxiety disappointment intensity satisfied appetite disgust interest skepticism approval disqualification scorn artificiality disregard involvement serenity astonishment disrespect joy servility at ease distress leniency shame attraction droopy loneliness sharpness balanced embarrassment longing shyness belonging embitterment love simplicity bitterness enjoyment meditative sincerity bliss envy mirth sleepy restlessness blur exaggeration misery slumber boldness excitement sorrow boredom fatigue naturalness stability calmness fear nervousness stubbornness caution firmness pain suffering clearness frankness panic superiority compassion fondness passiveness surprise complexity friendly patience suspiciousness concern frustration pity sympathy conciliated gaiety pleasure tenderness confidence generosity posing tension constraint gloom pride tolerance hate confusion grateful quiescence tranquility contempt greediness regret uneasiness contentment grievance relaxed unstable courage guilt relief vigilance yearning craving happiness repulsion weakness criticism haste respect worry curiosity “…emotion is a fact upon which all introspection agrees. [Most emotional states] are states which we have experienced personally. (Gellhorn & Loofbourrow ’63) Data of Emotion (Lang ’87) Everyone generally agrees on existence Basic datum is a state of feeling Completely private Include understanding of antecedents and consequences Important to determine how E is represented in memory Suggest a Turing test (but don’t describe…) Describing Emotion One approach: continuous dim. model (Cowie/Lang) Activation – evaluation space Add control Curse of dimensionality Primary E’s differ on at least 2 dimensions of this scale (Pereira) Computational Models (Pfeifer ’87) Emotion as process Emotion generation Influence of emotion Goal oriented nature Interaction between subsystems E. as heuristics Representation of emotion Computational Models (Pfeifer ’87) Examines models dimensionally A) Symbolic vs non-symbolic (cognitive vs AI) B) Augmented by emotion vs focused on emotion All approaches deal with E as process Unclear whether system state = emotion Models must function in complex, uncontrollable, unpredictable context No model for physiological aspect Emotions tightly coupled to commonsense reasoning Modeling Emotion in Speech Synthesis: basic issues (Schröder) How is a given emotion expressed? Which properties of the E state are to be expressed? Relationship between this state and another Approaches Formant synthesis (Burkhardt) Diphone concatenation Unit selection Modeling Emotion in Speech Formant synthesis (Burkhardt) High degree of control “emoSyn” Mean pitch, pitch range, variation, phrase and word contour, flutter, intensity, rate, phonation type, vowel precision, lip spread Two experiments Stimuli systematically varied, then classified Prototype generated and varied slightly Modeling Emotion in Speech Formant synthesis (Burkhardt) Fear High pitch, broad range, falsetto voice, fast rate Joy Broader pitch range, faster rate, modal or tense phonation, precise articulation Lowest recognition rates (perhaps due to intonation patterns) Boredom Lowered mean pitch, narrow range, slow rate, imprecise articulation Modeling Emotion in Speech Formant synthesis (Burkhardt) Sadness Narrow range, slow rate, breathy articulation Also raised pitch, falsetto Possible that sadness was imprecise term Anger Faster rate, tense phonation General results Recognition rates are comparable to natural speech, especially when the categories from experiment 2 are recombined. Modeling Emotion in Speech Generally: tradeoff between flexibility of modeling and naturalness: Rule-based less natural Selection-based is less flexible An Example – Ang ’02 Prosody-Based detection of annoyance/ frustration in human computer dialog DARPA Communicator Project Travel Planning Data (a simulation) (NIST, UC Boulder, CMU) Considers contributions of prosody, language model, and speaking style Doesn’t begin with a strong hypothesis An Example – Ang ’02 Uses recognizer output (sort of) Examines rel. of emotion and speaking style Uses hand coded style data Hyperaticulation, pauses, raised voice Repeated requests or corrections Hand labeled emotion relative to speaker Original and consensus labels An Example – Ang ’02 Emotion Class Instances Percent NEUTRAL 41545 83.84% ANNOYED 3777 7.62% FRUSTRATED 358 0.72% TIRED 328 0.66% AMUSED 326 0.66% OTHER 115 0.23% 3104 6.26% 49553 100.0% NOT-APPLICABLE Total An Example – Ang ’02 Prosodic Features Duration and speaking rate Pause, pitch, energy, spectral tilt Non-prosodic Features Repetitions & corrections Position in dialog Language model features Discriminated using decision trees “Brute force iterative algorithm” to determine useful features With and without LM features An Example – Ang ’02 Ang ’02 – Decision Tree Usage Temporal features 28% Longer duration, slow speaking rate corr. w/ frustration Pitch features 26% Generally, high F0 correlated w/ frustration Repeats/corrections (= system error) 26% Correlated w/ frustration Raised Voice Ang ’02 – Results Ang ’02 – Results Performance better by 5-6% for utterances on which labelers originally agreed Use of the repeat/correction feature improves success by 4% Frustration vs Else – very little data Only slight difference between labeled and recognized