SBIR-STTR Award

Because speech recognition R&D is conducted in large part through the creation and modification of computer software, productivity depends heavily of software engineering issues - the ease of writing new software and reusing existing software. These issues are relevant not lust to the work within a particular R&D group, but also to the exchange of technology among different groups. In particular, technology transfer among speech recognition groups is inhibited by the lack of convenient and powerful means for exchanging programs and data. To address this problem, a new speech recognition package (SRP) will be developed using advanced software engineering techniques, including abstract interfaces, object-oriented programming, and self-describing objects. These techniques have already been applied to speech and signal processing in the commercially-successful entropic signal processing system (ESPS) and waves+ (the ESPS graphics interface). ESPS and waves+ are rapidly becoming popular at many of the world is leading speech and signal processing centers. ESPS and waves+ will be used as a technology base for the SRP. Phase I will include a definition of requirements followed by the design and implementation of prototype modules. Anticipated benefits/potential commercial applications - the SRP will benefit the federal government because of the considerable speech recognition R&D it supports. The SRP will enable this work to proceed faster and more efficiently. Because the SRP will be developed as an extension to an existing commercial product already in use at speech recognition laboratories, the probability of commercial success is high.Key words: speech recognition, signal processing, information hiding, abstract types, modularity, object-oriented programming.

Because speech recognition R&D is conducted in large part through the creation and modification of computer software, productivity depends heavily on software engineering issues - the ease of writing new software and reusing existing software. These issues are relevant not just to the work within a particular R&D group, but also to the exchange of technology among different groups. In particular, technology transfer among speech recognition groups is inhibited by the lack of convenient and powerful means for exchanging programs and data. To address this problem, a new speech recognition package (SRP) will be developed using advanced software engineering techniques, including abstract interfaces, object oriented programming, and self-describing objects. These techniques have already been applied to speech and signal processing objects. These techniques have already been applied to speech and signal processing in the commercially-successful entropic signal processing system (ESPS) and *waves+* (the ESPS graphics interface). All of the Phase I technical objectives have been met. We have shown how ESPS programs can provide self-contained modules suitable for the SRP.We have developed a suitable software engineering methodology for the SRP. And we have demonstrated this methodology by implementing several low-level modules. The Phase II effort will complete the design and implementation of the SRP. Anticipated benefits/potential commercial applications - the speech recognition package (SRP) will benefit the federal government because of the considerable speech recognition and signal processing R&D it supports. The SRP will enable this work to proceed faster and more efficiently. Because the SRP will be developed as an extension to an existing commercial product already in use at speech recognition laboratories, the probability of commercial success is high.