1 paper
Shantanu Vispute, Aditya Mishra, Siddhartha Saxena
Speech transcripts used as long-term memory must preserve both words and stable speaker identities. Existing meeting-transcription metrics either ignore speakers or remap anonymous…