Glossary
Crosstalk
Overlapping speech where multiple people talk at once; common in natural conversation and challenging for ASR and diarization.
Overlap is constant in real conversation — backchannels, interruptions, excited agreement. A single-microphone recording mixes the overlapping voices into one waveform, which degrades transcription and inflates diarization error. Recording each speaker to a separate track preserves the overlap as usable signal while keeping every voice clean.
Why it matters
Multitrack recording turns crosstalk from a defect into training signal for exactly the cases models handle worst.