ISCA Archive IberSPEECH 2024
ISCA Archive IberSPEECH 2024

Fine-tuning Segmentation Models for the Albayzín diarization challenge

Mirari San Martín, Jónathan Heras, Gadea Mata

Speaker diarization aims to segment audio files according to different speakers and linking those segments which originate from the same speaker. Technologies for dealing with Speaker diarization can advance thanks to challenges like the Albayzín-RTVE 2024 Speaker Diarization and Identity Assignment Challenge, which provides labelled audios from Spanish TV shows. In this work, we have trained several segmentation models from the pyannotate.audio toolkit using the Diarizers library. Using this approach, we have build a diarization pipeline with a Diarization Error Rate (DER) of 0.2885 in our test partition of the Albayzín dataset.