Performance analysis for algorithms in noise cancellation

No Thumbnail Available
Marwa Yahia Qadous
Shereen Awni Bader
Journal Title
Journal ISSN
Volume Title
One of the most problems that affect almost all of our daily life is the background noise, which has many sources such as Transportation vehicles, aircraft, railroad stock, trucks, buses, automobiles, and motorcycles all producing excessive noise.These sources can be fall under some kind of noise; it may be white, colored or impulsive noise according to its power, and at what range of frequency it is occur.Since noise is unwanted and undesirable changes to our environment, needed arise to get rid of noise, and many attempts were held for this purpose, one of the most effective methods is the adaptive noise cancellation.Here we apply this method of adaptive noise cancellation on speech signal which has two main types voiced and unvoiced speech.An adaptive filter is a digital filter that can adjust its coefficients to give the best match to a given desired signal. When an adaptive filter operates in a changeable environment the filter coefficients can adapt in response to changes in the applied input signals. Adaptive filters depend on recursive algorithms to update their coefficients and train them to near the optimum solution.There are two main types of digital filters, finite impulse response (FIR) and infinite impulse response (IIR) filters. The filter output is calculated in a similar manner for both types.Also, we have many algorithms we can use to adjust the filter coefficients, such as least mean square LMS, normalized LMS, leaky LMS, recursive least square RLS which we discuss briefly in chapter 2.Using matlab we study and analyze the response of these algorithms as we change many parameters such as filter length and step size. These parameters affects convergence rate, MSE, computational complexity, number of iterations and stability, and all these are measures for the performance of our system of adaptive noise cancelation.Here we use two types of tests to compare between these algorithms and find the most effective one; objective and subjective tests.In objective test we use SNR and Segmental SNR as performance measure for this test; we examine the effect of changing language, speakers and gender on these algorithms.The subjective test which depends on listening to the output of our system, the input for the system is the noisy signal and the noise; the output should be the signal without noise.Finally we are going to compare between the results wither they matched or not