Speech enhancement using degenerate unmixing estimation technique and adaptive noise cancellation technique as a post signal processing

摘要

Speech enhancement in degenerate mixtures is a challenging task for the signal processing engineers. Degenerate Unmixing Estimation Technique (DUET) is a commonly used algorithm in speech signal separation, however there are limitations in separating the speech babble as noise and/or competing speech as noise from speech signal. These noises are important for auditory related training and testing. In this research an enhanced approach has been adopted by pairing DUET with Adaptive Noise Cancellation (ANC) which acts as the post-processor. Results from the correlation analysis and spectrogram analysis show that DUET provides separation quality correlation value of minimum 0.57 when speech signals are separated from speech babble and maximum of 0.75 when speech signal is separated from competing speech. On the contrary, both of these minimum and maximum correlation values reach to 0.77 and 0.88 when ANC is used as a postprocessor. The pairing of DUET with ANC has provided a good separation in the tested speech signals.