Finding speech instead of speaker: a model-based beamforming algorithm with heterogeneous ad-hoc microphone array for speech enhancement

Yang Zhang, Organizational Intelligence and Computational Social Science

ABSTRACT:
Imagine you are having a meeting in a conference room, where people have laptops and cell phones scattered all over the table. Is it possible to utilize microphones in these devices to enhance the speech? Traditional beamforming algorithms would say it’s hard because of the extreme challenge in locating the speaker under random configuration of the microphones. To solve that challenge, we proposed a new beamforming algorithm that does not rely on position calibration of the source at all. Instead, it maximizes the ‘speechness’ of the output signal guided by a strong speech model. Experiments show that the proposed algorithm produces cleaner and better sounding speech than traditional beamforming algorithms, and is immune to various problems in locating the speaker. In this talk we'll present the core concepts of the algorithm, and sample files recorded in real conference rooms.

Substrate Mechanics Influences Lung Endothelium Response to Thrombin Challenge

Arkaprava Dan, Department of Chemical and Biomolecular Engineering

ABSTRACT:
The endothelium is a critical component of the blood-gas barrier in the lung, permitting gaseous exchange while protecting against flooding of the airspaces by fluid edema from the capillaries. Thrombin, an inflammatory agent, triggers intracellular signaling cascades which causes failure of cell-cell junctions, resulting in gaps in the endothelial monolayer, compromising its barrier function. Aside from biochemical signals, the lung endothelium also receives mechanical signals such as substrate cyclic stretch during breathing, and substrate stiffening associated with disease and aging. We developed an equibiaxial stretcher capable of dynamically imaging lung endothelial monolayers subjected to cyclic stretch and different substrate stiffness, enabling us to track specific groups of cells over time as they responded to thrombin treatment. Quantification of gap areas in the monolayers revealed the dynamics of endothelial gap formation and closure, and established that physiological cyclic stretch and substrate stiffness contributed to the maintenance of monolayer integrity. The stretcher also enabled the visualization of monolayer recovery mechanisms such as lamellipodia protrusions, and their dependence on substrate mechanics. The results reveal the role of matrix stiffness in diseases such as pulmonary fibrosis where healthy tissue is replaced by stiffer scar tissue. We anticipate that the stretcher will be a valuable tool in investigating the stretch-mediated regulation of cell and tissue level functions.

Simultaneous Tri-Modal Measurement of Human Brain Function: Evidence for a Proof of Concept

Matt Moore, Social and Emotional Dimensions of Well-Being

ABSTRACT:
The link between spatial (where) and temporal (when) aspects of the neural correlates of most psychological phenomena is not clear. Elucidation of this relation requires integration across multiple brain imaging modalities and tasks that reliably modulate the engagement of brain systems of interest. The goal of this project is to illustrate the feasibility of such an integration across three imaging modalities (i.e., functional magnetic resonance imaging - fMRI, event-related optical signals - EROS, and electroencephalography/event-related potentials - EEG/ERP), by overcoming the methodological challenges posed by simultaneous recordings using these tools. This project provides initial evidence for a proof of concept, based on integrating data from pilot investigations using a cognitive task with emotional distraction. Results show that frontal dorso-ventral dissociations are effectively captured using fMRI, and similar dissociations are captured using EROS but in time windows similar to ERP responses, while the ERPs are captured at posterior locations. These results illuminate the overlaps and complementarity across measures of brain function, and point to ways in which distinct methodologies can be used simultaneously to investigate spatial and temporal correlates of psychophysiological phenomena.