Performance Analysis of Adaptive Noise Canceller in Digital Filter for Human Voice Separation

Author Name(s): Dr. Swati S. Godbole, Dr. Sanjay B. Pokle
Author Email: godboless@rknec.edu

Abstract

This Paper involves the performance analysis of Adaptive Noise Cancellation system (ANC) and its noise cancellation application by taking more realistic noises. Noise cancellation is a common occurrence in today’s telecommunication environment. The main purpose of this research is to eliminate unwarranted noise that exists in incoming signal, which makes it difficult to understand. Adaptive Noise Cancellation is an alternative technique of estimating signals corrupted by additive noise or interference. This paper examines adaptive filtering, employing discrete signal processing in MATLAB. Simulation was utilized by using MATLAB software to eliminate the noise. The strategies & design methodologies of Adaptive Noise Canceller using the least mean square (LMS) algorithm is considered in this paper. Using the LMS algorithm, adaptive noise canceller has been implemented on MATLAB successfully by considering more realistic noises for suitable filter order and convergence factor. The results are analyzed for different input noises originating from refrigerators, vaccum cleaners, hair dryers, washing machines as reference signals and speech signal corrupted by these noises as primary signal. These signals are tested and analyzed and the filter performance of these noises is measured in terms of signal to noise ratio (SNR) improvement. From the experimentation it is clear that signal to noise ratio of primary signal earlier it was 3.468db which gets substantially improved by 49.921dB. It has also been observed that for filter order N =50 and for convergence factor µ = 0.03, the voice signal can still be recovered perfectly.

Keywords Adaptive Noise Canceller (ANC), Least Mean Square (LMS) Algorithm, Signal to Noise Ratio (SNR)

Introduction

Acoustic problems have been in focus in the wake of technological growth that gave rise to undesirable noise emerge from different sources like household appliances to heavy machineries, sound systems etc. Such difficulties can sometimes trouble the neighbors next door. This method employ two different inputs, one is primary input that is information corrupted by noise & second reference input that is only noise [1]. For the processing of non-stationary signals adaptive filters are widely used & they are applied in different facets of communication. It consist of two main blocks, one is digital filter & second is adaptive evaluation. Corrupted information is effectively process by digital filter whereas algorithm is used for adaptation purpose by altering the coefficient & tries to optimize the desired weights [2-3].

Conclusion

From the results obtained, it can be concluded that the proposed ANC system has been successfully implemented on MATLAB by considering more realistic noises. The results are analyzed for different input noises originating from various disturbances such as construction site, air conditionings, refrigerators, barking of dogs, vaccum cleaners as reference signals and human voice signal as primary signal. These signals are analyzed and the filter performance is measured in terms of SNR improvement. It is evident from the results that SNR is clearly improved vis-à-vis filter performance. The speech signal is examined for different frequencies & voltage level to determine its capabilities of noise filtering. The first experiment is based on human voice where female speech is considered as original signal contaminated with vacuum cleaner noise signal recorded for different time durations. The second experiment is based on female speech signal contaminated with the vaccum cleaner noise recorded at t2 and the reference signal i.e. undesired vaccum cleaner noise with low speech recorded at t1. From the experimentation it is clear that SNR of primary signal earlier it was 3.468db which gets substantially improved by 49.921dB. Subsequent experiments also shows similar improvement in signal to noise ratio measured by considering more realistic noise signals. It has also been observed that for filter order N =50 and for convergence factor µ = 0.03, the voice signal can still be recovered perfectly. Based on these results, the designed system can effectively resolve the problem of unwanted noise and proved to be successful.

794 total views, no views today

Download PDF File

About the author: admin