Compressive Sensing (CS) was proposed as a promising compression technique at the sensing stage in a sensor for reducing size, cost and power of the sensor system. Recently a segmented approach was introduced at the sensing stage of CS and Kronecker-based technique was proposed for improving the recovery. This paper applies CS on windowed speech signals with 50% overlap and augments the performance of the Kronecker-based CS recovery technique by combining the overlapped part of the recovered windowed segments to obtain the recovered speech signal. Windowing of the speech signal reduces the spectral leakage while overlap reclaims the lost power due to windowing. The proposed improved method is tested on 8 female and 8 male speech signals from the TIMIT database. The proposed method achieved up to 14dB SNR improvement while recovering compressively sensed female and male speech over Standard CS recovery technique without overlap with a compression factor (CF) of 10%. The improvement reduced to 5dB at 50% CF.

Compressive Sensing, Kronecker Method, Reconstruction, Signal Quality, Speech, Window
2020 IEEE International Instrumentation and Measurement Technology Conference, I2MTC 2020
Department of Systems and Computer Engineering

Firouzeh, F.F. (Fereshteh Fakhar), Abdelazez, M. (Mohamed), Salsabili, S. (Sina), & Rajan, S. (2020). Improved recovery of compressive sensed speech. In I2MTC 2020 - International Instrumentation and Measurement Technology Conference, Proceedings. doi:10.1109/I2MTC43012.2020.9129262