Emotion Detection Overview
Vocea transmite mult mai mult decât cuvinte. Tonul, viteza, volumul dezvăluie starea emoțională. AI-ul poate detecta și reacționa în timp real.
Real-time
Frame-by-frame analysis
80%+
Average accuracy
6+
Emotions detected
Detectable Emotions
| Emotion | Voice Indicators | Accuracy |
|---|---|---|
| Neutral | Stable pitch, normal pace | 90%+ |
| Happy | Higher pitch, faster pace, rising intonation | 85% |
| Frustrated | Louder, faster, tense voice | 80% |
| Sad | Lower pitch, slower pace, monotone | 75% |
| Angry | Very loud, fast, aggressive tone | 85% |
| Confused | Hesitation, questioning intonation | 70% |
Acoustic Features
Prosodic
- • Pitch (F0) mean and variance
- • Speaking rate
- • Pause patterns
- • Intensity/loudness
Spectral
- • MFCC coefficients
- • Spectral centroid
- • Spectral flux
- • Formant frequencies
Voice Quality
- • Jitter (pitch variation)
- • Shimmer (amplitude variation)
- • HNR (harmonics-to-noise)
Processing Pipeline
// Real-time emotion detection
function analyzeEmotion(audioFrame) {
// Extract features
const features = {
pitch: extractPitch(audioFrame),
energy: calculateEnergy(audioFrame),
mfcc: extractMFCC(audioFrame),
speakingRate: estimateSpeakingRate(audioFrame)
};
// Normalize features
const normalized = normalizeFeatures(features);
// Run emotion classifier
const prediction = emotionModel.predict(normalized);
return {
emotion: prediction.label, // "frustrated"
confidence: prediction.score, // 0.85
valence: prediction.valence, // -0.6 (negative)
arousal: prediction.arousal // 0.8 (high energy)
};
}Applications
Escalation Detection
Auto-transfer când clientul e frustrat
Agent Coaching
Alert când tonul agentului e inadecvat
Sentiment Tracking
Monitor satisfacție pe parcursul apelului
Response Adaptation
AI adaptează tonul la emoția clientului