A research program is proposed to develop a searchable voice information system that will allow people who are blind or visually impaired to store and retrieve voice annotations from a large database of digitized audio recordings. The information retrieval system will be based upon an extension of word spotting technology that is extremely fast (over thirty hours of digitized audio can be searched in less than one second) and more accurate than speech recognition. Unlike speech recognition, Nexidia's word spotting performs a phonetic analysis of the search phrase and compares it to the phonetic patterns in a preprocessed digitized speech database. In addition, the word spotting methods continue to perform well in noisy, real world, environments where speech recognition algorithms routinely fail. With the proposed system, users will not need to perform structured data entry to store voice annotations, and they can quickly retrieve information by simply speaking a search phrase. The free format option for entering voice annotations is a significant advantage over all other competing audio-based information organization systems, and it will greatly simplify the user interface, making it immediately useful to the widest possible group of consumers. The proposed research will involve advanced digital speech processing, novel software interface design, and consumer involvement in the design and evaluation phases. The long-term goal is to create a software application for a personal mobile device that will increase independence and quality of life for people who are visually impaired. This is an assistive technology for rehabilitation, community re-entry, and independent living. The two specific aims for the program are: 1. Improve word spotting phrase detection with voice queries by employing same speaker model adaptation and secondary processing of query results. 2. Create a software interface that allows consumers who are blind or visually impaired to retrieve information more effectively than current methods of portable information management. Quantitative evaluation efforts will compute a Figure of Merit to measure how well the new word spotting algorithms perform for same and different speakers at different noise levels and for different noise types. Consumers who are visually impaired will test how well the system works compared to other methods currently used.
Thesaurus Terms: biomedical equipment development, blind aid, computer program /software, information system, portable biomedical equipment, voice bioengineering /biomedical engineering, clinical research, human subject
