Real-time separation of sound sources of spatial difference, by using combination of time space analysis and frequency space analysis (Final Year Project)
In the development of modern science, speech enhancement and recognition becomes more and more important in the world. In the recent decays, there appear huge number of speech enhance algorithms and software. Most of their works were mainly focus on deleting noise components while maintaining the main speech parts, which was so called “noise reduction”. Sometimes their assumptions were tedious and inadaptable: they often require that the noise component must maintain a series of assumptions, or the frequency of the noise must be different from that of the speech, or the speech part must occupy a large portion of energy. However, the real scene is, multiple people are talking at the same time while background music, noise are giving their performance simultaneously. Merely applying noise reduction methods will still result in mixed speeches and sounds. In this project, we mainly focus on real-time blind source separation. We aim to separate each source component from mixed sources via Delayed Signal Subtraction with Time-Frequency Masking in a Real Time manner, so as to make a practical speech enhance system.
Keywords: Sound Separation, Microphone Arrays, Delayed Signal Subtraction.