Speaker: Juan Ignacio Álvarez Trejos.
Abstract:
This talk presents the results of our participation in the speaker diarization challenge at IberSpeech2024. Our approach combines the strengths of three diarization models: a custom-trained Diaper model, Pyannote, and VBx, through an innovative fusion strategy enhanced with doverlap. Beyond standard diarization error rate (DER) analysis, we delve into additional metrics such as speaker counting error and speaker turn detection, offering a more comprehensive evaluation of our system’s performance. This deeper metric analysis highlights the robustness and areas for improvement of our approach, providing valuable insights into the state-of-the-art in speaker diarization.