Randomized trial analyzes tonal structure in political talk shows, suggesting insights into discourse dynamics.
Key Points
The research aims to develop a computational framework to analyze tonal structure and dynamics in political discussions on Turkish television.
Utilized a dataset of televised debates featuring journalists, politicians, and commentators.
Trained a convolutional neural network (CNN) on Mel-frequency cepstral coefficients (MFCCs) to classify speech segments into sociable, discuss, and monotone categories.
Employed ridge regression with a CLR transformation to model speaker influence on tonal composition.
Achieved an average classification accuracy of 92.3% across five-fold cross-validation.
Utilized entropy-based metrics to analyze tonal diversity and balance across episodes.
Identified significant patterns in tonal trajectories and segment transitions during episodes of political talk shows.