Speech Noise Clustering



Clustering is a major task in machine learning and many different clustering techniques are available. Very small academical effort was invested in finding the intersection between machine learning algorithms and digital signal processing (DSP) tasks. This research tried to utilize clustering algorithms, in particular spectral clustering and Independent Component Analysis, to reduce noise from speech centric audio recordings. The results were compared with a noise-reduced outcome, which was generated using existing post-production techniques, such as equalizing and frequency bandpasses, that were implemented using Python. The noise-reduced audio outcome using machine learning algorithms was not as good as the audio outcome using post-production techniques, but led to some interesting conclusions and ideas for further research.