SBIR-STTR Award

This Small Business Innovation Research Phase I project by OmniSpeech LLC will address the need to have clear communications in noisy environments. Cellphone users want to make calls in noisy situations, and the inability to have clear communication in noise limits when and where phone calls can be made. Omnispeech is developing a completely software-based single-microphone solution which works in stationary and non-stationary noisy environments. It also has the potential to clean up the speech of the cellphone user when it is corrupted by the speech of another competing talker. During Phase I, Omnispeech will address the problem of distinguishing noisy unvoiced consonants from background noise. Another major focus will be optimizing the speech extraction algorithm so it can run in real-time on a DSP processor. Although, the proposed technology has immense potential for (1) Voice over IP, (2) 1st responder market, (3) assistive technologies for the hearing impaired, and (4) military/intelligence/homeland security, to begin with, Omnispeech will focus on the mobile industry. There were 1.1 billion mobile units worldwide in 2009 and global mobile phone sales for 2010 are anticipated to increase by 9%. Sales of smartphones in 2009 comprised 14% of overall mobile handset sales representing approximately 165 million units, a 24% increase from 2008. The number of new smartphone and web-enabled devices sold in 2013 is estimated to be 674 million units and that number is predicted to climb in subsequent years. OmniSpeech will target smartphones initially. In the following year, OmniSpeech will launch into feature phones. These two segments of the market, if properly addressed, will provide OmniSpeech a clear path to significant revenue which can be used to grow into the other application areas mentioned above.

This Small Business Innovation Research (SBIR) Phase II project addresses the problem of everyday noisy environments that limit when and where people can be heard clearly over various communication devices (phones, first-responder radios, voice-over-IP, etc.). Given the inability of single-microphone noise-reduction techniques to handle non-stationary noise (e.g. restaurant noise) the industry's current solution is the use of dual-microphone techniques. However, dual-microphone techniques cost more, require more hardware, need spatial separation between the noise and speech, and can only process the signal on the transmit side. The proposed technology is a single-microphone software-only solution that effectively handles non-stationary noise coming from any direction and can process the signal on both the transmit and receive sides. The novelty of the approach is its use of speech-specific characteristics and knowledge of human perception to extract speech from the noisy signal. The objective of the proposed research is to improve the current method of detecting voice activity, enabling better speech extraction thereby resulting in enhanced speech quality. The resulting technology will be architecture agnostic, cost effective and have superior performance in everyday situations. The broader impact/commercial potential of this project is considerable and compelling. The initial focus will be the mobile phone industry, which is now the single largest user of noise suppression products with a market size that includes nearly all of humanity. The technology will then be optimized to improve the listening experience for hundreds of millions of potential hearing-aid/cochlear implant users worldwide. According to the World Health Organization, there are 278 million people worldwide that have hearing loss. This number is expected to at least double over the next 30 years due to the growth in the number of senior citizens over 65 years of age and the growth in the number of younger people who are needing hearing aids 20 years sooner than their parents due to loud-music listening habits. Additional markets include first-responder radios, voice-over-IP and military/intelligence/homeland-security. The research required to reduce computational complexity will illuminate the essential aspects of auditory scene analysis needed for improved speech perception