16th Annual Conference of the International Speech Communication Association

Dresden, Germany
September 6-10, 2015

Model-Based Adaptive Pre-Processing of Speech for Enhanced Intelligibility in Noise and Reverberation

Jan Rennies (1), Andreas Volgenandt (1), Henning Schepker (2), Simon Doclo (1)

(1) Fraunhofer IDMT, Germany
(2) Carl von Ossietzky Universität Oldenburg, Germany

In this demonstrator we present the most recent advances in the development of the near-end listening enhancement algorithm AdaptDRC. The algorithm uses short-time estimates of the speech intelligibility index to control spectral shaping, dynamic range compression and/or level adjustment to achieve an adaptive enhancement of speech intelligibility in adverse listening conditions. Depending on the application scenario, the algorithm framework can take background noise and reverberation as well as different boundary conditions into account. The show and tell contribution comprises a real-time setup of the algorithm to demonstrate the sound modifications and the impact of the different parameters and boundary conditions. An accompanying poster shows results of formal listening tests evaluating the speech intelligibility improvement achieved by the algorithm for normal-hearing and hearing-impaired listeners.

Full Paper

Bibliographic reference.  Rennies, Jan / Volgenandt, Andreas / Schepker, Henning / Doclo, Simon (2015): "Model-based adaptive pre-processing of speech for enhanced intelligibility in noise and reverberation", In INTERSPEECH-2015, 2619-2620.