Frame Identity - Multi-Modal Biometric Verification

Behavioral and biometric identity verification using voice, typing, and face recognition.

Extracted from the Frame microservices architecture.

Features

• Voice Recognition: Biometric voice pattern matching with embedding similarity
• Typing Dynamics: Behavioral keystroke pattern analysis (timing, pressure, rhythm)
• Face Recognition: Facial biometric verification (future)
• Multimodal Fusion: Bayesian combination of multiple biometrics
• Trust Integration: Bridge to multi-dimensional trust system
• Pattern Learning: Adaptive baseline refinement over time

Quick Start

[dependencies]
sam-identity = "0.1.0"

Dependency Architecture

frame-identity depends on:

frame-identity
├── frame-catalog (database, embeddings)
└── frame-sentinel (trust scoring)

Used by: Frame core for biometric authentication

Position in Frame ecosystem:

frame-catalog ──┬→ frame-identity
                │
frame-sentinel ─┘

use sam_identity::{VoiceStore, TypingPatternStore};
use frame_catalog::Database;

// Initialize stores
let db = Database::new("identity.db")?;
let voice_store = VoiceStore::new(&db)?;
let typing_store = TypingPatternStore::new(&db)?;

// Voice authentication
voice_store.store_voice_sample("user123", &voice_embedding)?;
let similarity = voice_store.verify_voice("user123", &query_embedding)?;
println!("Voice similarity: {:.2}%", similarity * 100.0);

// Typing pattern analysis
let pattern = TypingPattern::from_keystrokes(&keystroke_data);
typing_store.store_pattern("user123", &pattern)?;
let match_score = typing_store.verify_pattern("user123", &pattern)?;

Modules

• identity (380 LOC) - User identity management
• typing_patterns (551 LOC) - Keystroke dynamics analysis
• voice_store (289 LOC) - Voice biometric storage
• face_store (175 LOC) - Face biometric storage (future)
• multimodal (400 LOC) - Multi-modal fusion
• typing_trust_bridge (63 LOC) - Typing → trust integration
• voice_trust_bridge (50 LOC) - Voice → trust integration

Voice Recognition

• Store voice embeddings (384-dim vectors)
• Cosine similarity matching
• Baseline refinement with moving average
• Similarity threshold: 0.85 for verification

Typing Dynamics

Pattern Features:

• Key hold time (dwell time)
• Flight time (key-to-key interval)
• Typing rhythm variance
• Error correction patterns
• Common n-gram timings

Analysis:

• Statistical baseline per user
• Deviation scoring
• Pattern drift detection

Multimodal Fusion

Combine multiple biometric modalities:

use sam_identity::{MultiModalIdentifier, Modality, ModalityConfidence};

let identifier = MultiModalIdentifier::new();

// Add modality scores
let confidences = vec![
    ModalityConfidence {
        modality: Modality::Voice,
        user_id: "user123".to_string(),
        confidence: 0.92,
    },
    ModalityConfidence {
        modality: Modality::Typing,
        user_id: "user123".to_string(),
        confidence: 0.85,
    },
];

let result = identifier.identify(&confidences)?;
println!("Verified: {} (confidence: {:.2})", result.user_id, result.confidence);

Trust Integration

Bridge biometric scores to multi-dimensional trust:

use sam_identity::{VoiceTrustBridge, TypingTrustBridge};

// Voice → Trust
let voice_bridge = VoiceTrustBridge::new(voice_store, trust_manager);
voice_bridge.update_voice_trust("user123", &voice_embedding)?;

// Typing → Trust
let typing_bridge = TypingTrustBridge::new(typing_store, trust_manager);
typing_bridge.update_typing_trust("user123", &keystroke_pattern)?;

Compatibility

• Rust Edition: 2021
• MSRV: 1.70+
• Platforms: All

Dependencies

• frame-catalog - Database, embeddings
• frame-sentinel - Multi-dimensional trust
• rusqlite (0.31) - Persistence
• bincode (1.3) - Binary serialization
• regex (1.10) - Pattern matching

License

MIT - See LICENSE for details.

Author

Magnus Trent magnus@blackfall.dev

Links

• GitHub: https://github.com/Blackfall-Labs/sam-identity