Robust Automatic Transcription of Speech–Transition (RATS-T)
SRI is improving the robustness of speech processing technologies and transitioning them to government partners for use in highly noisy operational audio conditions.
Current speech signal processing technologies often cannot adequately process noisy or highly degraded speech in scenarios that are important to government and military personnel. This can mean that a significant amount of time is dedicated to analyzing distorted audio recordings. The Defense Advanced Research Projects Agency (DARPA) Robust Automatic Transcription of Speech (RATS) program aims to create algorithms and software that can automatically extract critical information from speech-containing signals received over extremely noisy and/or highly distorted communication channels.
Under the RATS program, SRI's Speech Technology and Research (STAR) Laboratory developed speech processing algorithms and software for speech activity detection (SAD), language identification (LID), speaker identification (SID), and keyword spotting (KWS) over distorted communication channels.
STAR Laboratory researchers focused on high-accuracy, robust, fast, and field-ready automatic speech processing systems for operational conditions. Since error rates were reduced beyond expectations and the resulting systems showed high utility in operational use cases, SRI was awarded a sole-source contract as a follow-on phase called RATS–Transition (RATS-T).
The primary goal of RATS-T is to investigate new techniques for adapting speech processing algorithms once deployed so that performance is not degraded on unseen conditions—i.e., that were not observed during system development. A second objective is to improve the engineering, efficiency, and speed of the software to prepare it for transition to field use. SRI is working directly with transition partners in the government.
Distribution Statement "A": Approved for Public Release, Distribution Unlimited