Speech recognition using multiple microphones is used for multiple commercial and medical applications. Currently, speech processing is processed at each channel independently. However, availability of the multiple microphones enables use of the spatial information on the speaker position.
The main goal of this project is development of the acoustic beamforming methods to use the spatial information on the speaker position for speech recognition improvement.
The Project includes:
-
Development of the acoustic beamforming methods for speaker localization
-
Development of DNN approach for multi-channel speech processing using spatial information using MATLAB and C
-
Using the microphone array for the acoustic data collection
-
Train the DNN algorithms using the collected data
-
Demonstrate the performance improvement of the developed multi-channel approach comparing to the conventional methods
The project will be performed under joint supervision by Dr. Yaniv Zigel (Email: yaniv@bgu.ac.il) and Dr. Igal Bilik (bilik@bgu.ac.il ).