DiarizationLM: Speaker Diarization Post-Processing with Large Language Models

Name: DiarizationLM: Speaker Diarization Post-Processing with Large Language Models
Uploaded: 2026-10-15T20:08:32.798602+00:00
Duration: 1129 s
Description: DiarizationLM uses large language models to improve speaker diarization outputs by post-processing them. It combines outputs from existing systems, uses a prompt builder, and leverages LLM finetuning techniques like hyp2ora to drastically reduce word diarization error rates on datasets like Fisher and Callhome.

DiarizationLM uses large language models to improve speaker diarization outputs by post-processing them. It combines outputs from existing systems, uses a prompt builder, and leverages LLM finetuning techniques like hyp2ora to drastically reduce word diarization error rates on datasets like Fisher and Callhome.

Open the full episode