Recently by the increasing use of digital media especially multimedia archives, researchers have been looking for safe reliable and easy to use methods by means of these Medias. There have been different methods developed in which among them speaker diarization can be pinpointed. In the latter method, vocal tract is one of the features which has been taken into account. Since the structure of larynx changes rapidly during a talk or conversation, more stable features can be used for speaker identification purposes. In this paper we tried to use lower parts of larynx known as SGR (Sub Glottal Resonance), as features in the segmentation stage of speaker diarization. After simulations and analysing the results segmentation and diarization phases, the proposed method has given 63 percent of accuracy in the first phase and 91.67 percent in the second phase.
Paper
Full text
Speaker Diarization in Broadcast News Using Sub Glottal Resonances
Semantic Scholar · Computer Science · 2019
Abstract
Recently by the increasing use of digital media especially multimedia archives, researchers have been looking for safe reliable and easy to use methods by means of these Medias. There have been different methods developed in which among them speaker diarization can be pinpointed. In the latter method, vocal tract is one of the features which has been taken into account. Since the structure of larynx changes rapidly during a talk or conversation, more stable features can be used for speaker identification purposes. In this paper we tried to use lower parts of larynx known as SGR (Sub Glottal Resonance), as features in the segmentation stage of speaker diarization. After simulations and analysing the results segmentation and diarization phases, the proposed method has given 63 percent of accuracy in the first phase and 91.67 percent in the second phase.