With its multidisciplinary mix of skilled engineers, computer scientists, and linguists, SRI’s Speech Technology and Research (STAR) Laboratory creates and transfers speech technology for government and commercial clients. We build systems that meet client needs for speech recognition, speaker and language identification, speech translation, human/computer dialog systems, accurate recognition in noisy environments, wordspotting, language education, speech analytics, and more.
Research and Development Areas
Our work includes
- Speech recognition: Automatic speech recognition for multiple languages and genres; speech-to-speech translation; keyword spotting; speech production and perception-based features; noise robustness; prosodic modeling; disfluencies
- Natural language understanding: Multilingual information extraction/retrieval; human-computer interaction; dialog systems; error detection and recovery; virtual personal assistant (VPA) technology; semantic analysis; syntactic parsing
- Machine translation: Translation of many spoken/written language styles; information retrieval; machine-mediated human-to-human cross-lingual communication
- Information extraction: Identifying topics and events; question answers; summaries; high-level annotations
- Speech analytics and speaker/audio characterization: Voice biometrics/speaker identification; speech/language-based demographic detection; language/accent identification; emotion/affect detection; mental health assessment; social role modeling; audio event detection; speaker diarization
Under the Defense Advanced Research Project Agency (DARPA) Robust Automatic Transcription of Speech (RATS) program, the STAR Laboratory is breaking new ground with algorithms and software for speech activity detection, language identification, speaker identification, and keyword spotting in noisy signals.
With funding from the National Science Foundation, the STAR Lab and SRI's Center for Technology in Learning are working to find ways to assess collaboration within a group of learners by applying automatic speech processing. More information
The STAR Lab is developing technology for DARPA's Broad Operational Language Translation (BOLT) program. The project's goal is to translate foreign languages in all genres, retrieve information from translated material, and enable bilingual communication.
For DARPA's Active Authentication program, we developed novel ways to validate the identity of a person interacting with a phone or computer by looking at characteristics of their spoken and written language – to reduce the need to remember long, complex passwords. More information
SRI’s SenSay Analytics™ real-time platform for speaker state analytics will soon be available for licensing. The platform is being used for research and development in multiple projects on speaker state detection from speech, including emotion, cognition, health, and mental health.