We show technology for spoken-document retrieval -- a search engine for finding speech recordings in a large collection of audio files. We can also find videos by searching their sound track.
It can be used for online lectures and training, archives of video-taped presentations, and recorded meetings and tele-conferences.
Unlike most such systems, our technology has unlimited vocabulary -- we have no problem finding uncommon words such as people names, special terminology, or code names. Using our novel phonetic indexing technology, we are able to search hundreds of hours of video in less than a second.

This is a real-time algorithm to render the rich visual effects of general non-height-field geometric details, known as mesostructure. Our method is based on a five-dimensional generalized displacement map (GDM) that represents the distance of solid mesostructure along any ray cast from any point within a volumetric sample. With this GDM information, we propose a technique that computes mesostructure visibility jointly in object space and texture space which enables both control of texture distortion and efficient computation of texture coordinates and shadowing. GDM can be rendered with either local or global illumination as a per-pixel process in graphics hardware to achieve real-time rendering of general mesostructure.

The demo illustrates the ShareNETS project, running in the wireless and networking group, MSRA. The ShareNETS project focuses on end-systems and tries to leverage all these end systems to cooperatively form a self-organized cognitive network. The end-system based network can extend and enhance the existing infrastructure-based networks in many cases. Multiple services can also be supported in such an end-system based network. The demo shows a scenario that provides VoIP service in ShareNETS over a multi-radio, multi-hop wireless network. In the demo, two colleagues are calling each other using VoIP service. The call is initiated when one part is connected on Ethernet. Then, in order to catch a meeting, the guy disconnects the Ethernet and continues to get connected with the other part via multi-hop wireless network. Our global roaming support enables the VoIP session seamlessly migrated to the new network interface and the conversation is not interrupted. Moreover, ShareNETS enables real-time service (like VoIP) in a multi-hop wireless environment using conventional hardware. To do that, we develop 2.5 Layer MAC, which is software enhanced MAC layer that has better support of QoS. Based on this, joint routing and reservation are also designed for end-to-end coordination and resource allocation for real-time services.

Most modern computers or game consoles are equipped with powerful yet cost-effective graphics processing units (GPUs) to accelerate graphics operations. Though the graphics engines in these GPUs are specially designed for graphics operations, can we harness their computing power for more general non-graphics operations? The answer is positive. This project is focusing on the GPU acceleration of generic video decoding. Specifically, we achieved real time playing back of WMV9 coded High-Definition video on Xbox which has an Intel Pentium III 733 MHz CPU and a nVidia GeForce 3 GPU.

As a key feature of Microsoft’s Tablet PC, digital ink is a human-centric technology that provides for a rich and natural representation of human input. It combines the expressive power of human handwriting with the power of a computer. To make digital ink a first-class data type in Microsoft Windows, MSR Asia has focused on: 1. Smart ink analysis (parsing, detection and recognition); and 2. Novel ink user interface (editing and manipulation). With these technologies, unstructured notes are turned structured and daily tasks such as note-taking and editing can be managed entirely in the ink domain, using a natural and friendly user interface. These technologies give users the complete freedom in writing, diagramming, sketching, and capturing creative ideas using a pen.

English writing wizard (EWW) is an English proofing system that helps ESL (English as a Second Language) users in word choice, or more specifically, the use of collocation. Collocation is a habitual usage of words. For example, we would say “I had a cup of strong tea” (in stead of “I had a cup of powerful tea”) and “I made a plan” (instead of “I did a plan”). Since the use of collocation is unpredictable by grammars, it is an ‘intermediate plateau’ where many English learners remain stuck.
This demo shows the preliminary research results, including collection error detection and correction, and collocation dictionary look up.

In contrast with the current Web search methods that essentially do document-level ranking and retrieval, we are exploring a new paradigm to enable Web search at the object level. Specifically, we extract and integrate all the related Web information about the same object together as an information unit called a Web object, and rank these Web objects in terms of their relevance and popularity to answer a user query.
Libra (http://research.microsoft.com/Libra) is a testbed for evaluating the techniques we developed for searching the Web at the object level, and a free computer science bibliography search engine. You can use Libra paper search engine to:
1.Find top scientists, conferences, and journals in your field (try Author Search, Conference Search…);
2.See how research communities emerge and evolve (try Interest Group search);
3.Locate papers you are interested in with better ranking;
4.Identify rising stars or hot papers in your field (try Advanced Search)
The engine currently covers 1 million papers.

DigiParty is a research prototype for multiparty audio/video communication. It supports audio/video conversations among up to five persons. DigiParty leverages our application-level multicast technology - DigiMetro, and is able to provide the best audio/video experience under limited bandwidth conditions.
DigiMetro is an application-level multicast subway tailored to small and impromptu video conferencing. Breaking through the conventional wisdom to use shared overlay to handle multiple data sources, DigiMetro organizes the data delivery routes as source-specific trees, which are first constructed by a local greedy algorithm and then gradually improved by a global refinement procedure. Extensive simulation experiments demonstrate the efficiency of both algorithms. Moreover, DigiMetro is able to handle different video bit rates and provides different services over voice/video streams.

Studies have shown that digital photos are most frequently viewed by the person who took the photo. When photos are shown to others, the photographer is usually physically present, directing and narrating the showing. Despite this, most current photo user interfaces are design for either annotating and organizing photos or for producing photo artifacts, such as DVDs, slideshows, or web sites.
PhotoPie! is a user interface specifically designed for showing photos – either to yourself or others. It is extremely simple, scales to thousands of images, and does not require the user to do any organization. Other interfaces are too cumbersome or complicated to use during live photo narration, but the PhotoPie! is so simple that the operator does not even need to look at the controls. PhotoPie! works equally well on a desktop machine, a TV, a PocketPC, and a SmartPhone.
PhotoPie! takes advantage of the most recent usability findings for photo narration. With digital cameras, users are taking hundreds of photos on their vacations. The previous techniques of simply showing one photo after another are inadequate for such a large number of images. However, it is not a matter of selecting a small number of “good” images either, as all of the images are of approximately equal quality. The PhotoPie! interface allows users to easily view clusters of several images at once. The clusters are formed automatically by looking for statistically significant differences in the time the images were taken. The user can then narrate an entire cluster at a time, or can zoom in to break a cluster into smaller and smaller subclusters, to the level of individual images.

Lazy Snapping is an interactive image cutout tool. Lazy Snapping separates coarse and fine scale processing, making object specification and detailed adjustment easy. Moreover, Lazy Snapping provides instant visual feedback, snapping the cutout contour to the true object boundary efficiently despite the presence of ambiguous or low contrast edges. Instant feedback is made possible by a novel image segmentation algorithm which combines graph cut with pre-computed over-segmentation. A set of intuitive user interface (UI) tools is designed and implemented to provide flexible control and editing for the users. Usability studies indicate that Lazy Snapping provides a better user experience and produces better segmentation results than the state-of-the-art interactive image cutout tool, Magnetic Lasso in Adobe Photoshop.