The Effect Of Musical Noise Pro Duced By Ibm

956 Words Sep 18th, 2016 4 Pages
verberant mixture. In 16 a minimum mean-square error (MMSE) based technique is planned to estimate the ideal multi-threshold mask (IMM) that has been utilized in the realm of monaural speech improvement. It contains 2 stages, specifically training stage and enhancement stage. Within the training stage, a man-made neural network is trained by exploitation the SNR of every T-F unit of training information. Second stage uses the calculated SNR to estimate IMM and to separate the target speech from clamant signal.

This analysis work proposes a method to scale back the impact of musical noise pro-duced by IBM, by planning a soft mask which might be utilized in speech separation applications. Genetic algorithmic rule (GA) is employed during this work to search out the optimum soft mask weights between 0 and 1. the objective measures like S/N improvement and perceptual evaluation of speech quality (PESQ) are used to measure the performance of planned optimum soft mask with the prevailing IBM 22 and IMM16 based speech separation systems. Rest of the paper is organized in the following manner. Section two provides an outline of computational auditory scene analysis (CASA). Section three presents the proposed optimum soft mask based speech separation system. Section four provides the experimental results of IBM, IMM and also the planned soft mask. Section five deals with the conclusion and future work.

2. Computational Auditory Scene Analysis (CASA)

CASA can be defined as the…
Open Document