Red Shift Company

Red Shift Company, LLC (“Red Shift”) is a Delaware Limited Liability Company formed to create and commercialize a revolutionary new line of voice products built on bio-mimicry principles that model human hearing and speech in a completely new way. Red Shift believes its products will be significantly superior to the traditional commercial products of Automatic Speech Recognition (“ASR”), Automated Text-To-Speech (“TTS”), Hearing Aids, Noise Cancellation (reduction), Voice Identification, Speaker and Microphone technologies, and all human-machine speech interfaces.

The Red Shift bioscience discovery made by John F. Remillard, President of Red Shift, is a radical departure from the traditional beliefs and understandings of the biology of hearing and speech. The original and unique nature of this radical approach increases the chances of being issued patents that are less likely to be challenged. Red Shift’s techniques are very radical improvements that do not conform or conflict with traditional speech processing techniques.

ASR Noise Enhancement

Red Shift enhances speech signal processing by considering the mechanics of the auditory cortex and the human speech generation mechanism from the viewpoint of biomimicry. In particular, ASR is done based on biologically salient components of speech. Red Shift improves word recognition accuracy for ASR by extracting traditional features for these engines using proprietary biological assumptions rather than assumptions usually made when transforming stationary signals to the frequency domain.

The value of this technology can be quantified by simply generating new features for ASR. To date, Red Shift has shown over 60% improvement of Cambridge University's Hidden Markov Model ToolKit (Microsoft copyright) and over 50% improvement of Carnegie Mellon University's Sphinx 3 decoder (for both continuous and semi-continuous models). These results beat the accuracy baseline of LumenVox - a commercially distributed speech engine with a noise canceling front end. Furthermore, when the LumenVox noise cancellation is first applied to the test signal, the Red Shift Feature Extractor still provides 30%-40% improvement. In other words, the Red Shift Feature Extractor provides an additive enhancement to commercial grade noise cancellation.

The results for the Red Shift Feature Extractor are published here for industry standard tests in two regimes: digits in noise and large vocabulary with noise added. For ASR, it is crucial to explore noise-robust features that maintain high performance at various signal-to-noise ratios (SNRs). Machines do about as well as humans for low noise and as bad at high noise but poorly in comparison for the mid-range SNRs of 15 dB to 10 dB. As Red Shift technology comes from modeling the cochlea, it may not come as a surprise that we find this gap is greatly closed.

Finally, Red Shift has filed 5 domestic patent applications and 3 international patent applications to protect this technology.