Combination of binaural Wiener filtering noise reduction and blind source separation for automatic speech recognition
In real life scenarios, speech must often be recognized in noisy environments, where the target speech is contaminated by both noise and interfering speech. Applications that require speech recognition (teleconferencing, automatic speech recognition (ASR), hearing aids, etc) do not work well in such environments. Traditional speech enhancement algorithms often work only in narrowly specified conditions or with specific noise statistics. Algorithms exist and work well when the background noise is stationary and non-speech; however, these algorithms often fail when competing speakers are present. A possible solution to this problem is to use source separation algorithms like beamforming to enhance the target speech. However, beamforming algorithms require a priori knowledge about the acoustic environment and the sources involved, or a large number of sensors are required for good performance. Another algorithm for source separation is blind source separation (BSS). BSS estimates sources only based on the information in signals observed at each input channel; it requires no a priori knowledge and furthermore requires only a small number of microphones. The aim of our mini project is to develop a framework for dealing with a given realistic like acoustical environment that was defined in the PASCAL 'CHiME' challenge. some most recent noise PSD tracking algorithms, and find a suitable procedure for evaluating the performance of noise estimators.
