Alternative Transducers for Handheld Automatic Speech Recognition in Military Environments

The proposal describes unique and promising solutions for robust speech recognition in handheld devices, particularly for the DARPA Phraselator, which was developed under a previous SBIR and a rapid development program. The proposed solutions are: (1) anew special designed microphone array with directional microphones and corresponding algorithms, such as beam forming, adaptive noise cancellation, and passive source localization to improve the quality of input voice signals; and (2) a new auditory-basedfeature extraction algorithm to improve speech recognition in noisy environments. These two solutions can be applied separately or together to construct an alternative transducer/new front-end, to replace the traditional transducer system. The goal of thisproject is to apply the results of our speech recognition technology to develop more efficient and effective transducers/front end to enhance multilingual automatic speech recognition and translation in the Phraselator and other handheld devices. Ashandheld devices and wireless communications gain more acceptance and wide usage, there will be a significant and sizeable market for the technology developed in this project. First, almost all the PDA's on the market today have a voice recording function.In noisy environment, users have to put the PDA's very closely to their mouth to get acceptable recoding quality. Using the microphone array and the new algorithms developed in this project, the quality of the recorded voice will be improved significantlyunder the same recording condition; Second, the same technology can be applied to wireless handset; therefore, the quality of the voice will be improved while the users are talking in a moving car or on streets with background noise; and Last, there areseveral companies which have already implemented automatic speech recognition and translation technology in PDA's and wireless handsets, and they provide market for the improved solutions developed under this SBIR project. The technology developed in thisproject can be used directly to improve the performances of speech recognition and translation in the handheld devices. Furthermore, we expect the solutions to be available in a very rapid development cycle and thus support pressing DoD needs. Also, weanticipate the proposed solutions will provide significant improvement with minimum added cost penalty. Upon the above analysis, Li Creative Technologies will be benefited significantly from licensing the new technologies developed from this project to thewireless and PDA industries and providing supports and further improvements continuously.