Crowdsourcing Audio Semantics by Means of Hybrid Bimodal Segmentation with Hierarchical Classification

Vrysis L, Tsipas N, Dimoulas C, Papanikolaou G. Crowdsourcing Audio Semantics by Means of Hybrid Bimodal Segmentation with Hierarchical Classification. J Audio Eng Soc. 2016;64(12):1042-1054. doi:10.17743/jaes.2016.0051

@article{Vrysis2016_18537,
  author = {Vrysis, Lazaros and Tsipas, Nikolaos and Dimoulas, Charalampos and Papanikolaou, George},
  title = {{Crowdsourcing Audio Semantics by Means of Hybrid Bimodal Segmentation with Hierarchical Classification}},
  journal = {Journal of the Audio Engineering Society},
  volume = {64},
  number = {12},
  pages = {1042--1054},
  year = {2016},
  month = dec,
  publisher = {Audio Engineering Society},
  doi = {10.17743/jaes.2016.0051},
  url = {https://doi.org/10.17743/jaes.2016.0051}
}

TY  - JOUR
TI  - Crowdsourcing Audio Semantics by Means of Hybrid Bimodal Segmentation with Hierarchical Classification
AU  - Vrysis, Lazaros
AU  - Tsipas, Nikolaos
AU  - Dimoulas, Charalampos
AU  - Papanikolaou, George
T2  - Journal of the Audio Engineering Society
J2  - J. Audio Eng. Soc.
VL  - 64
IS  - 12
SP  - 1042
EP  - 1054
PY  - 2016
DA  - 2016/12/06
DO  - 10.17743/jaes.2016.0051
UR  - https://doi.org/10.17743/jaes.2016.0051
PB  - Audio Engineering Society
LA  - en
AB  - The task of general audio detection and segmentation is quite common in contemporary audio applications where computationally intensive processes are frequently involved. Machine learning is usually employed along with user-enabled data labeling that is intended to detect, segment, and semantically annotate the relevant audio events. This work focuses on a generic audio detection and classification method that combines hierarchical bimodal segmentation with hybrid pattern classification at different temporal resolutions. This paper presents the algorithmic perspective of a mobile back-end system to facilitate the construction, validation, and continuous update of generic audio ground-truth data. The goal is the implementation of a system that is capable of performing well in different conditions without relying on complicated pattern recognition systems and taxonomies. For this reason, minimal prior knowledge is necessary so that there is consistent behavior for different input signals and computational environments. Novel “classification confidence” metrics are implemented.
ER  -