Yuanyuan Zu

  • Citations Per Year
Learn More
Using the computational auditory scene analysis (CASA) as a framework, a novel speech separation approach based on dual-microphone energy difference (DMED) is proposed for close-talk system. The energy levels of the two microphones are calculated in time-frequency (T-F) units. The DMEDs are calculated as the energy level ratio between the two microphones,(More)
The difficulty of active hearing protection system is to mask the dangerous noise energy and retain the useful signal at the same time. Binary mask, which was an effective method to be used in speech segregation, was proposed to micro active hearing protection system. A new gammatone filterbank approximation with less resource requirement was proposed to(More)
This paper addresses the problem of close talk speech enhancement as a binary classification using dual microphones features in noisy and reverberant environments. In this work, we investigate a speech segregation framework, in which deep neural networks (DNN) are employed as a mechanism to find the robustness classifier from two microphones inputs. The(More)
The speech segregation and enhancement is a hard task in speech communication. In order to get the clean target speech, a close talk system is used to collect the speech with a nearby microphone. A deep neural networks (DNN) estimator is used in a frequency channel for speech energy calculation with parameter masks. The adjusted binaural auditory features(More)
The reverberant speech segregation is a basic problem in speech enhancement and automatic speech recognition. Based on the deep neural networks (DNN), a novel binaural speech segregation method is proposed. The binaural feature is extracted and used as the cue to train a DNN with a ideal parameter mask. The trained DNN is used to distinguish the target(More)
Based on MRI technology, a series of 2D images of human head was obtained. Their boundary coordinates were extracted respectively, and a hierarchical 3D head parameterized model was constructed by fitting contours of all sections with Fourier series. Several feature sections were defined according to main features of human face. Average processing was taken(More)
  • 1