SPINE: Speech in Noisy Environments

Investigators

Ramana Rao Gadde
Harry Bratt
Horacio Franco
Martin Graciarena
Kemal Sönmez
Andreas Stolcke
Anand Venkataraman
Dimitra Vergyri
Jing Zheng

Project Summary

SRI participated regularly in the evaluations of speech recognition systems on speech collected under noisy conditions. The SPINE task was sponsored by DARPA.

SRI's SPINE recognition systems is based on the DECIPHER recognizer and builds on the research carried out under the Consistency Modeling and LVCSR projects.

One of our research projects, funded by DARPA's Reliable Omni-Present Automatic Recognition (ROAR) program, involves the use of multiple sensors for robustness in noisy environments.

Another line of research under this program explores approach to dialog modeling without large amounts of labeled training data.

Publications and Presentations

• V.R.R. Gadde & A. Stolcke, The SRI Spine 2000 Evaluation System, Evaluation of Speech Technology in Noisy Environments Workshop, Naval Research Laboratory, Washington, DC, 10-11 October 2000.
• V.R.R. Gadde, A. Stolcke, D. Vergyri, J. Zheng, K. Sonmez, & A. Venkataraman, SRI 2001 SPINE Evaluation System, Presentation at the DARPA SPINE Workshop, Orlando, FL, Nov. 29, 2001. (PowerPoint)
• H. Franco, M. Graciarena, K. Sonmez, & H. Bratt, Combining Heterogeneous Sensors with Standard Microphones for Noise Robust Recognition, Presentation at the DARPA ROAR Workshop, Orlando, FL, Nov. 30, 2001. (PowerPoint)
• V. R. Rao Gadde, A. Stolcke, D. Vergyri, J. Zheng, K. Sonmez, and A. Venkataraman (2002), Building an ASR System for Noisy Environments: SRI's 2001 SPINE Evaluation System. Proc. Intl. Conf. on Spoken Language Processing, Denver, vol. 3, pp. 1577-1580. (PDF)
• A. Venkataraman, A. Stolcke, & E. Shriberg (2002), Automatic Dialog Act Tagging with Minimal Supervision. Proc. 9th Australian International Conference on Speech Science and Technology, Melbourne. (PDF)
• A. Venkataraman, L. Ferrer, A. Stolcke, & E. Shriberg (2003), Training a Prosody-based Dialog Act Tagger from Unlabeled Data. To appear in Proc. IEEE Intl. Conf. on Acoustics, Speech and Signal Processing, Hong Kong. (PDF)