Analyzing Candidate Speaking Time in Estonian Parliament Election Debates
Siim Talts, Tanel Alumäe · Digital Humanities in the Nordic and Baltic Countries Publications · 2020
In this paper, we analyze the amount of speaking time by each candidate and political party during the election debates that aired in broadcast media during the Estonian 2019 parliament election campaign, using automatic speaker identification. We use automated methods for retrieving speech recordings from publicly available sources that are likely to contain speech by the target speakers, and apply a weakly supervised method for training speaker recognition models from such data. The experiments show that the resulting models have high precision but they lack in recall. While most candidates from large and established parties could be identified automatically, candidates from smaller and newer parties could not be recognized, due to their little prior media presence. Our system allows to identify candidates who spoke for the longest time in the election debates. The analysis shows that the election debates were not biased from the speaking time point of view: all major political parties received relatively similar speaking time across the debates.