Improved Speaker Diarization & Persistent Speaker Tagging

Speaker diarization currently struggles with accurately identifying speakers, especially in single-speaker conversations where the system should confidently recognize one speaker at all times. Additionally, users should have the ability to tag speakers manually, and those tags should persist across conversations, reducing the need for repeated identification.

Challenges

• Inconsistent speaker recognition: A user speaking solo should be identified with 100% accuracy.

• Limited diarization accuracy: The current model (Deepgram’s diarization) misidentifies or fails to consistently assign speaker labels.

• Lack of persistent speaker tags: Users must repeatedly tag speakers, as the system does not remember identities across different conversations.