🟡 [MEDIUM] Build cross-meeting speaker matching

[ucguy4u]

Description

Implement similarity scoring to recognize same speaker across different meetings.

Tasks

• Implement cosine similarity for voice embeddings
• Add confidence thresholds (e.g., >0.85 = same person)
• Handle edge cases (similar voices, background noise)
• Add speaker matching API
• Test with multiple meetings

Estimated Time

2 days

Priority

MEDIUM - Enables persona continuity

Dependencies

• Voice embedding generation complete

Related Files

• src/persona/speaker_matcher.py
• src/persona/voice_memory.py