Perception-Based Automatic Classification of Background Noise in Super-Wideband Telephony

Tiemounou S, Jeannès RLB, Ewert VB. Perception-Based Automatic Classification of Background Noise in Super-Wideband Telephony. J Audio Eng Soc. 2014;62(11):776-781. doi:10.17743/jaes.2014.0040

@article{Tiemounou2014_17552,
  author = {Tiemounou, Sibiri and Jeannès, Régine Le Bouquin and Ewert, Vincent Barriac},
  title = {{Perception-Based Automatic Classification of Background Noise in Super-Wideband Telephony}},
  journal = {Journal of the Audio Engineering Society},
  volume = {62},
  number = {11},
  pages = {776--781},
  year = {2014},
  month = nov,
  publisher = {Audio Engineering Society},
  doi = {10.17743/jaes.2014.0040},
  url = {https://doi.org/10.17743/jaes.2014.0040}
}

TY  - JOUR
TI  - Perception-Based Automatic Classification of Background Noise in Super-Wideband Telephony
AU  - Tiemounou, Sibiri
AU  - Jeannès, Régine Le Bouquin
AU  - Ewert, Vincent Barriac
T2  - Journal of the Audio Engineering Society
J2  - J. Audio Eng. Soc.
VL  - 62
IS  - 11
SP  - 776
EP  - 781
PY  - 2014
DA  - 2014/11/06
DO  - 10.17743/jaes.2014.0040
UR  - https://doi.org/10.17743/jaes.2014.0040
PB  - Audio Engineering Society
LA  - en
AB  - Given the effort required to perform subjective tests of quality of service in super-wideband telephone networks, an objective model of perceived degradation was created. This paper describes an efficient classification model of background noise arising at the originating side of super-wideband telephony. Classification depends on the impact on the perceived voice quality at receiving side from originating noises. This approach extends the research findings of narrowband telephony. Subjective experiments showed that background noises can be divided into three relevant classes: environmental noise, breath noise, and crackling noise. Each of these corresponds respectively to spectral flow, acoustic power variations, and the noise spectral centroid. Based on three quality degradation indicators, tests showed 82% correct classification of unknown noise signals.
ER  -