Engine Cards

What It Analyzes

• Micro-tremor patterns associated with cognitive load during deception
• Acted vs. genuine speech classification (trained on CREMA-D + VoxPopuli)
• Vocal stress indicators (fundamental frequency variation, jitter, shimmer)
• Temporal consistency and speech rate anomalies

Scientific Basis

• Ekman & Friesen (1969) - Micro-expression and vocal leakage theory
• DePaulo et al. (2003) - Cues to deception meta-analysis
• Kirchhubel & Howard (2013) - Acoustic correlates of deceptive speech

Known Limitations

• No lie detector is 100% accurate. Results are probabilistic, not deterministic.
• Cultural differences affect vocal expression patterns significantly.
• Professional actors or trained individuals may confound detection.
• Audio quality (background noise, compression) reduces accuracy.

What It Analyzes

• Big Five personality traits inferred from vocal patterns (openness, conscientiousness, extraversion, agreeableness, neuroticism)
• Emotional valence and arousal from prosodic features
• Communication style classification (assertive, empathetic, analytical, expressive)
• 42-emotion intelligence matrix from acoustic biomarkers

Scientific Basis

• Scherer (2003) - Vocal communication of emotion
• Mohammadi & Vinciarelli (2012) - Automatic personality perception from speech
• Mairesse et al. (2007) - Using linguistic cues for personality recognition

Known Limitations

• Personality inference from voice is probabilistic; self-report questionnaires remain the gold standard.
• Emotional state at recording time affects trait inference.
• Accents and dialects may influence perceived traits.

What It Analyzes

• Vocal stress biomarkers (cortisol-correlated frequency patterns)
• Cognitive load estimation from speech disfluencies and timing
• Vocal fatigue and strain indicators
• Respiratory pattern analysis from speech segments

Scientific Basis

• Titze (1994) - Principles of Voice Production
• Cummins et al. (2015) - Speech analysis for health applications
• Low et al. (2020) - Automated assessment of psychiatric disorders using speech

Known Limitations

• NOT a medical device. Results do not constitute medical diagnosis.
• Clinical indicators are correlational, not causational.
• Environmental factors (noise, microphone quality) affect readings.
• Individual baseline variation is significant; longitudinal tracking improves accuracy.

What It Analyzes

• Authority and confidence indicators in speech patterns
• Persuasion effectiveness metrics (pace, emphasis, pause strategy)
• Communication clarity and articulation scoring
• Vocal presence and projection assessment

Known Limitations

• Professional voice assessment is context-dependent; a courtroom voice differs from a coaching voice.
• Cultural norms affect what constitutes "authoritative" or "persuasive" speech.
• Should be used as coaching feedback, not employment screening.

What It Analyzes

• Fundamental frequency (F0), formants, harmonics-to-noise ratio
• MFCC features (40 coefficients + deltas) for spectral characterization
• Temporal features: speech rate, pause patterns, rhythm
• Voice quality: breathiness, roughness, strain, vocal age estimation

Technical Details

• Audio input: WAV/MP3/WebM, 8kHz-48kHz sampling rate
• Feature extraction: librosa, praat-parselmouth
• ML architecture: BiLSTM + Transformer attention (custom ORAVYS models)
• Inference: ONNX Runtime for CPU-optimized real-time analysis