Speech recognition and separation research are highly susceptible to noise and overlapping sounds in multi-speaker situations. A target extraction method is a valuable front-end solution for later procedures such as classification and interpretation. Natural speech is continuous since it contains both overlapping and non-overlapping voices. Since overlapping voices produce a complicated combination of signals, many approaches are applied to overcome this task. This study investigates a speech separation model to improve the flexibility of speaker separation in intricate, multi-speaker overlapping circumstances. In this work, we presented speech separation of mixed-signal. Initially, we calculated the Mel Frequency Cepstral Coefficients (MFCC) features to convert the data in the time-frequency domain, and the inverse of MFCC is calculated to convert the signals back into the time domain. The aim of the proposed work is to improve the flexibility of both the speaker identification and speech separation models in intricate, multi-speaker, overlapping circumstances. Metrics like subjective score or signal-to-noise ratio and the precision of subsequent tasks like speaker identification can be used to assess speech separation quality.