Two-Microphone Speech Enhancement Using a Learned Binary Mask
Keywords:
Noise reduction, Binary mask, speech/noiseclassification, Two-microphone features.
Abstract:
Ideal binary mask speech enhancement is shown toincrease the speech quality as well as speech intelligibility. But,this property depends highly on the accurate separation ofspeech and masker time-frequency units of the input spectrum,which is a difficult task in real situations. Ordinary binary maskmethods are single-microphone methods and so, can obtainlittle information from the environment. In this paper, we devisea two-microphone method that uses a classifier to distinguishspeech-dominated and masker-dominated time-frequency units. The classifier uses simply computable two-microphone featureswhich enable it to be used in real-time scenarios. Theseproposed features empower the classifier to reach toclassification accuracies near 80%. This high accuracy in turn,empowers the Ideal binary mask mthod to obtain higher SNRIand NPLR values in comparison to state-of-the-art noisereduction methods. These results indicate that the proposedtwo-microphone features have high information content forspeech/masker separation.