@inproceedings {Kollmeier1719_2016,
	year = {2016},
	author = {Kollmeier, Birger and Lenarz, Thomas and Warzybok, Anna and Schädler, Marc and Haumann, Sabine and Brand, Thomas and Lücke, Jörg},
	title = {Auditory Profile and Common Audiological Functional Parameters (CAFPAs): From Diagnostics to Machine-Learning-based Evidence},
	booktitle = {Assoc. Res. Otolaryng. MidWinter Meeting (ARO)},
	URL = {http://c.ymcdn.com/sites/www.aro.org/resource/resmgr/Abstract_Archives/UPDATED_2016_ARO_Abstract_Bo.pdf},
	abstract = {# Introduction # How  well  should  the  various  audiological  findings  best be  represented  and  how  can  this  information  be  used  to  characterize  the  individual  hearing  problem  of  each  patient  –  preferably  in  a  way  which  is  independent  from  his  or  her  native language? This contribution reviews the approach and models  developed  in  the  Cluster  of  Excellence  Hearing4All  (Oldenburg/Hannover)  to  unite  the  diverse  audiological databases,   e.g.   from   the   hearing   research   and   clinical   institutions in Oldenburg and in Hannover in a more abstract, comprehensive  way  than  the  previously  defined  “auditory profiles”. # Method # A set of “common audiological functional parameters” (CAFPAs) has been defined that serves as an abstract representation of  the  most  important  audiological  characteristics  of  each  patient. The CAFPAs include, e.g., sensitivity loss in different frequency  regions,  distortion  component  and  compression  loss  at low  and  high  frequencies,  central  and  binaural  loss,  cognitive and socio-economic component of the hearing loss. These  CAFPAs  have  been  defined  in  order  to  make  non-consistent and non-complete audiological data accessible for methods of machine learning, such as, e.g. Bayesian nets. Since speech recognition tests in noise are the most crucial outcome parameter, the multilingual matrix test (Kollmeier et al., 2015 Int. J. Audiol. online first) is used which is suitable not only for comparisons across different clinics but also across languages.  The  individual  Speech  Recognition  Thresholds  (SRT) in stationary and in fluctuating noise were predicted using  the  audiogram  and  an  estimate  of  the  internal  level  uncertainty  as  parameters  following  the  Automatic  Speech  Recognition (ASR) approach by Schädler et al. (2015, Int. J. Audiol. Online first)
# Results # Using  such  statistical  methods,  a  data-driven  audiological  classification for different classes of hearing loss becomes possible.   Estimates   of   the   “typical”   hearing   loss   and   suprathreshold  distortion  components  in  combination  with  ASR-based  speech  recognition  prediction  allow  to  predict  the individual performance for the closed-set Matrix sentence recognition test in different languages. 
# Conclusions # A  consistency  check  across  the  different  audiological  input  and outcome measures becomes possible by using auditory models  adapted  to  the  individual  hearing  impairment.  The  concept  of  a  more  abstract  representation  of  audiological  diagnostical    information    in    combination    with    speech    recognition  prediction  methods  and  other  machine  learning  approaches  appears  to  be  promising  for  further  research  in  diagnostical and rehabilitative audiology.
# Funding# DFG, EXC 1077 Hearing4All}
}