Successfully developed, enhanced and optimized signal and decoding algorithms used to automatically transcribe speech into text across multiple languages. Modeled, tested, designed, and optimized algorithms for speed, accuracy, and scalability. Led the enhancement and optimization of SAIC’s PlainSpeech ASR decoder to include several new features. (C/C++)
• Designed uniform front-end features from different engines.
• Implemented linear discrimination methods, language model look-ahead method, key/hot word recognition, and test-to-speech alignment
• Enhanced the finite state automata (FSA) functionalities.
• Developed SRGS-compatible grammar-based speech recognition to support voice based 511 systems.
• Optimized ASR engine to dynamically manage lexicons, language models, and grammar.
• Improved the ability of the engine to perform real time recognition.
• Significantly reduced CPU time and memory requirements, while improving speech recognition accuracy.
• Designed and developed a high quality multi-pass speech recognition architecture.
• Led the design and implementation of the SAIC Advanced Speech Recognition System for IVR solutions with a grammar parser that fully supports the W3C Speech Recognition Grammar Specification (SRGS).