Prototypes and Demos
- News Video Summarization System
This is a live news video indexing and search system that incorporates a multimodal approach to multimedia search. This was achieved by exploiting all available information from the inherent audio (e.g.
ambient sound, speech), video (e.g. moving images, on-screen OCR-text) and external textual information (e.g. news scripts, web articles). All meta-data extraction, done with I2R's patent-pending technology, was performed automatically. The system has been running for nearly 1 year (as of August 2007), monitoring 7 Mediacorp news channels daily.
Please click the following links to go to the demo pages of the system:
- Joint Source-Channel Model for Machine Transliteration
The joint source-channel model is a new paradigm for bidirectional proper name transliteration between the latin-alphabetic languages and the CJK (Chinese, Japanese and Korean) languages. It jointly models the knowledge source and the information channels to reflect the close coupling between languages. The new paradigm achieves a quantum leap improvement in accuracy upon the state-of-the-art transliteration methods. The model can be easily generalized for machine translation as well.
- EVITA : I2R Extensible Voice Portal
EVITA (Extensible Enterprise Voice Portal) is the enterprise portal of the EVITA project's ready-made packaged voice application suite, for voice-enabling the enterprise. The initial suite comprises of ADA, My ADA, Common Directories, Personal Phonebook Manager, and will include HR Helpdesk, Customer Helpdesk, etc, as well as other applications for a voice-enabled enterprise. Future extensions will include packaged vertical industry applications (financial, travel, healthcare, insurance, etc).
- EVITA-RAD : Web-based VUI Rapid Application Development
Speech interactivity and call automation technology has reached a point of reliability sufficient for deployment at the enterprise level. Even though Voice User Interface (VUI) applications are growing in popularity, the development process remains a bottleneck to its widespread adoption. Currently available development tools are meant for either professionals or programmers experienced in VoiceXML, speech recognition, dialogue design, and platform specific languages. We present a web-based tool, EVITA-RAD, driven by an extensible and modular design paradigm for building dialogues. By incorporating database access and abstracting many issues of programming in VoiceXML into modules, our tool allows users to focus on designing dynamic and useful dialogue applications. Thus, the EVITA-RAD lowers the barrier of entry for VUI application development, and helps promote widespread adoption of VUI applications.
|
|
|
