SBIR-STTR Award

The objective is the exploratory research to parameterize the speech signal and provide measurements which maximize the speech intelligibility and talker-identification while minimizing the effects of noise and interference. Performance of speech transducers and noise canceling systems have been bench-marked significantly for applications under controlled conditions. This basic research will investigate new signal processing and speech parameterization techniques for more robust performance in noise conditions. Additionally, the effects of channel media interference will be quantified and superimposed for several expected applications to evaluate the enhanced performance derived from these new concepts. The new algorithms and noise/interference abatement techniques will demonstrate immunity to a broad range of acoustic noise and transmission media conditions. The technical approach consists of two parallel activities which individually focus on: 1) the identification of novel ways to isolate the signal from the noise to improve speech intelligibility and speaker identification scores, and 2) development of performance bench-marks to demonstrate the improvements in speech recognition under noisy acoustic conditions and over channels that filter or otherwise interfere with speech signal quality features.

Phase II will make use of the performance benchmark and noise characterization tools developed in Phase I to accomplish the proposed testing of different speech capture appliances and medical monitoring transducers for telemedicine application. These articulation index (AI) performance and Fast Fourier Transform (FFT) characterization tools will benchmark the various remedies giving repeatable and reliable data during the proposed field trials. Field trials will use the automated FFT and performance benchmark algorithms in a portable and self-contained device. TERI will fabricate several prototype devices to gather the benchmark data in highly mobile situations. These devices incorporate the automated algorithm "sensing" capability and, in real-time modify the signal capture parameters to optimally adjust to environmental noise characteristics. Tasks proposed for the Phase II research continues the successful speech capture experimentation and culminates in a product for commercialization in Phase III. The technical approach consists of: !) Completing the automated benchmark testing in various noise environments; 2) Investigating the physical appliance portion of the speech capture system; 3) Fabricating self-contained prototypes suitable for field testing; 4) Conducting field trial tests with the prototypes; and 5) Refining the speech capture system design for miniaturization and commercialization.