audio/video research

* Media services for 3G telephony: Transferred video-streaming capabilities (H.263, MP4 video, AMR) to HP's Open Call Business Unit. Provided interactive VCR controls to an unmodified SIP client while complying with 3G and 324M-gateway session contraints, including "seek" and speed controls. Listed as one of top-40 innovations from the 40-year history of HP Labs * Streaming-server modeling: Modeled performance of streaming media service, based on "black-box" measurements. Approach used a small number of distinct workload types for calibration. Combinations of constrained total-least-squares models successfully modeled server saturation points for mixed workloads for Real's Helix Server and Apple's Darwin Server, including non-linearities for the saturation surface.

very-low-bit-rate video messaging

* Dissolve and fade detection * Automatic selection of highlight images ("snapshots") from video * Faster-than-real-time MPEG1 video analysis * Automatic placement of DVD chapter points * Computationally efficient approaches for low-resolution audio alignment * Automatic detection of stable and distinctive audio segments for matching * Automatic music-tempo estimation * Motion-tolerant de-interlacing of MPEG2 (without explicit motion estimation) * "Snow static" and other forms of video "garbage" detection

In 2000, the patents covering on my work here were valued at $17 to $22 million in an independent audit by KPMG International. * Speeding up audio in the MPEG-compressed domain in a computationally and memory efficient way without pitch shifting. * Speeding up speech while maintaining comprehensibility by mimicking natural timing of fast speech. * Estimating articulated-figure pose based on the silhouette without convolution. * Tracking articulated-figure motions using video-rate illumination and depth imagery. * Generating automatically photo-realistic lip sync video ("Video Rewrite") * Locating specific points on objects using convolutional offset estimation and linear subspace coupling. * Tracking object boundaries across variable backgrounds * Morphing images and audio automatically without detailed manual alignment. * Skimming and segmentation of video * Estimating personalized head-related transfer functions

acoustic noise control