Noise PSD Estimation in reverberant environments for speech intelligibility enhancement algorithms
Topic: Intelligibility enhancement algorithms can be used to increase the speech intelligibility of speech processing algorithms like public address systems and hands free systems. These algorithms depend on the noise PSD in the environment of the user. Any reverberation of the target signal that leaks into the PSD estimate will degrade the intelligibility of target signal. In This project we will study the development of noise PSD estimation algorithms in reverberant environments.
Intelligibility enhancement algorithms can be used to increase the speech intelligibility of speech processing algorithms like public address systems, hands free systems, hearing aids, etc. With these algorithms, the target signal is pre-processed before play-back by the loudspeaker, such that the speech intelligibility of the target as received by the user is maximised. These algorithms typically depend on the power spectral density (PSD) of the acoustic noise in the environment of the user. However, typically, the target signal gets reverberated after playback in the environment of the user. Any reverberation that leaks into the noise PSD estimate will degrade the final intelligibility as received by the user. In this project we will develop an algorithm that can be used to estimate the noise PSD for speech intelligibility enhancement algorithms in noisy reverberant environments.
Project outside the universityIn this project it is possible to collaborate with our industrial partner Bosch security systems B.v.
Assignment- Implementation of an existing speech intelligibility enhancement algorithm in matlab/simulink - development of a noise PSD estimation algorithm meant to be used in noisy reverberant environments.
dr.ir. Richard Hendriks
Circuits and Systems Group
Department of Microelectronics
Last modified: 2017-01-27