Abstract :

In this work, we offer an improved mask-based noise reduction strategy for improving voice signals. First, the Amplitude Magnitude Spectrogram is determined by decomposing the noisy speech signal into its component time-frequency (TF) components. (AMS). The signals are then sorted into groups according to their quality ratio to provide a preliminary set of answers. The Cuckoo search technique is then used to produce the best mask for each class. The waveform synthesis step next involves windowing the filtered waveforms, multiplying them by the best mask value, and summing the products to get the final, improved target signal. Experiments were run on many datasets to gauge the suggested method's efficacy, and the results were compared to those of established methods using signal-to-noise ratio. The acquired findings validated the efficiency of the suggested method in reducing background noise and improving the quality of the spoken signal.