In this project wavelet transform has been used for reduction in noisy speech signals. The wavelet transform should give good results when applied to removal. The wavelet filter is represented in the frequency and time domains. The use of different wavelets and different orders has been evaluated for their suitability as a speech noise removal. Two types of wavelets have been used, these being the Biorthogonal wavelets and Daubechies wavelets. The measurement used is Error Metric. The input speeches are sampled, at 10, 30, and 100 seconds; both male and female voices are used. The Daubechies wavelets perform better than Biorthogonal wavelets. The average value of Error Metric for Daubechies at 5 levels with a Wiener filter is 0.962 and is the best result. The Discrete Wavelet Transform performs well for the elimination of noise in speech data.
