Use a speech-to-text system that supports speaker diarization
Separate audio sources before transcription when possible
Apply source separation to isolate overlapping voices
Transcribe each separated stream independently
Align transcripts with timestamps to reconstruct the conversation
Mark overlapping segments explicitly in the transcript
Label speakers consistently across the full recording
Manually review ambiguous overlap sections
Use higher-quality audio recordings to improve separation
Reduce background noise before transcription
Slow down playback for difficult overlap segments
Combine automated transcription with human correction
Use short transcription windows to capture rapid turn-taking
Preserve partial words when speakers interrupt each other
Annotate unintelligible overlap where words cannot be recovered