Automated Recognition of Sincere Apologies from Acoustics of Speech

Sincerity is an important characteristic of communicative behavior which represents an honest, truthful, and genuine display of verbal and non-verbal expressions. Individuals who are deemed sincere often appear more charismatic and can influence a large number of people. In this paper, we propose a multi-model fusion framework to identify sincerely delivered apologies by modelling difference between acoustics of sincere and insincere utterances. The efficacy of this framework is benchmarked using the Sincere Apology Corpus (SAC). We show that our proposed methods can improve the baseline classification performance (in terms of unweighted average recall) for SAC from 66.02% to 70.97% for the validation partition and 66.61% to 75.49% for the test partition. Moreover, as part of our investigation, we found that gender dependency can influence the classification performance of machine learning models, with models trained for male subjects performing better than those trained for female subjects.

@article{Syed2020,
  title     = {Automated Recognition of Sincere Apologies from Acoustics of Speech},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {11},
  number    = {6},
  year      = {2020},
  publisher = {The Science and Information Organization},
  author    = {Zafi Sherhan Syed and Muhammad Shehram Shah and Abbas Shah Syed},
  doi       = {10.14569/IJACSA.2020.0110679},
  url       = {https://doi.org/10.14569/IJACSA.2020.0110679}
}