• Prof. Pooja Rasane
• Harshada Bhujbal
• Omkar Dhore
• Maithili Jagdale
• Sandhyarani Sonkamble

In recent years, deep learning-based methods have significantly improved the performance of speech separation. However, a challenging problem remains: how to separate the speech of unknown speakers in a mixture, a problem known as speaker-independent speech separation. This research presents an innovative deep learning framework for speech separation that addresses the issue of unknown speakers in the mixture. We propose a neural network that projects the time-frequency representation of the mixture signal into a high-dimensional feature space. Within this space, reference points (attractors) are created to represent each speaker, defined as the centroid of the speaker in the embedding space. The time-frequency embeddings of each speaker are then encouraged to cluster around their corresponding attractor points, which are used to determine the time-frequency assignment of each speaker. This approach enhances the robustness of speech separation for various applications, including speech recognition, speaker diarization, and more.

Speech Separation, DNN, Feature Extraction, Deep Learning, Speaker-Independent.

"Speaker-Independant Speech Separation With Deep Attractor Network", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.11, Issue 4, page no.e693-e698, April-2024, Available :http://www.jetir.org/papers/JETIR2404475.pdf

"Speaker-Independant Speech Separation With Deep Attractor Network", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.11, Issue 4, page no. ppe693-e698, April-2024, Available at : http://www.jetir.org/papers/JETIR2404475.pdf