Tom Gajecki and Waldo Nogueira presented the work "Development of a Sound Coding Strategy based on a Deep Recurrent Neural Network for Monaural Source Separation in Cochlear Implants" at the ITG 2016 Speech Communication in Padeborn, Germany. This work proposes a cochlear implant sound coding strategy arquitecture based on a deep recurrent neural network (DRNN) to enhance a target speaker in the presence of a competing voice. The algorithm has been evaluated in normal hearing listeners using a VoCoder and cochlear implant users. it seems that the algorithm provides larger improvements in sppech intelligibility for normal hearing listenersthan for cochlear implant users. The hypothesis to explain these contradictory results may be that cochlear implant users are more sensitive to the distortions introduced by the DRNN in the target signal than normal hearing listerns using the VoCoder.
The poster and the manuscript can be obtained from these links ([poster] [paper]). Some Pictures of Tom Gajecki presenting the poster are given below.
22 Comments
|
Archives
July 2021
Categories |