THE USE OF AI IN MULTITALKER SPEECH SEPARATION: RelaJet Enables Hard of Hearing People to Hear Clearly Again

22 Jan

Hard of hearing people face two main problems. Firstly, hearing aids are expensive. Take the top six global brands for example: their average price is about $2000, while high-end pieces could cost as much as $5000. Secondly, many conventional hearing aids are not effective enough because they are susceptible to interference from ambient noise. Blue Chen, a hard of hearing person himself, has gathered an AI algorithms team to found RelaJet. He aims to employ “the multi-talker speech separation” to enhance the listening experience for hard of hearing people.

Blue Chen’s Team Helps Hard of Hearing People to Zero in on a Particular Voice

Any problem in the auditory hierarchy, from sound reception in the human ear to obtaining meaningful interpretation in the brain, can be regarded as hearing impairment. Hearing aids usually pick up background noise making it difficult for hard of hearing people to monitor the primary sound. On the other hand, the normal human auditory system can focus on the primary sounds automatically filtering out background noise and other irrelevant conversations. For instance, at a noisy party, a normal person can pay attention to the speech of nearby friends and hear someone hollering at a distance as well. As long as one concentrates on the primary sound, the ambient noise would feel three times less loud. This process is termed the “cocktail party effect.” Overall, RelaJet aims to assist hard of hearing people to improve their detection, discrimination and identification in the auditory hierarchy. To enable a hard of hearing person to better concentrate on the primary sound, one must first find the special characteristics. Through RelaJet’s neural network engine and by utilizing deep learning, one can complete multitalker speech separation within 10 milliseconds. There are two main ways to achieve this objective. The first approach involves “pre-recorded sound eigenvalues.” Suppose Jimmy needs to converse with Ellen in a noisy room, Jimmy could first ask Ellen to record 2 to 3 seconds of her voice in his smartphone app. After the multi-talker speech separation has noted the eigenvalues of Ellen’s speech, the associated microphone would henceforth strengthen her voice while minimizing the ambient noise. The second is the so-called “ABX blind listening test.” The smartphone app does not pre-specify the eigenvalues of anybody’s speech. Instead, the multi-talker speech separation would instantaneously determine how many people are speaking, and then the user would select the eigenvalues of the desired speaking partner’s sound. The multi-talker speech separation will strengthen the desired sound while eliminating background noise.


Empathy for the Hard of Hearing People Fosters Innovation under the Most Trying Circumstances

Both sound input and output are encoded via pulse-code modulation (PCM). Achieving discrimination and identification of distinct eigenvalues within 10 milliseconds represents multitalker speak separation’s greatest advantage. Why 10 milliseconds? That is because if hearing aids complete voice processing beyond this limit, the human ear will sense this delay which can cause dizziness. Hence, for a hearing aid to be classified as a medical device, it must finish all speech processing within just 10 milliseconds.

Chen points to his own hearing aids and says, “I have been lucky. A hearing aid piece costs around $4,500 each—meaning $9,000 for both ears. Although I can afford hearing aids, I still could not match normal people’s hearing and pronunciation, much less those who could not afford hearing aids.” Helping less fortunate patients has been the overarching motivation for the founding of RelaJet. As such, Chen demands that his products undergo the same rigorous medical worthiness review as other medical devices on the market. As a first step, he will introduce this technology to the top six global makers of hearing aids. Chen believes that if he could gain enough market share in the healthcare industry, then branching into the markets for Bluetooth headsets, smart speakers and other products with lower consumer demand would become easier.

▲ As the co-founder of RelaJet and also a hard of hearing person, Blue Chen can provide valuable feedback to accelerate RelaJet’s R&D process.

RelaJet Closely Monitors Policy and Market Shifts and Will Use Licensing to Gain Market Share

When asked why he resigned from his job at MediaTek to become an entrepreneur at this time, Chen states that he anticipates a seismic shift in the US hearing aid market. In 2020, the US Food and Drug Administration (FDA) will legally allow Over-the-Counter (OTC) hearing aids to come on the market. Future hearing aids will no longer undergo the time-consuming process to obtain medical certification. In the future, an AI algorithm can be used to certify hearing aids, significantly reducing the cost of these devices and making them more widely available.

Besides the FDA-certified hearing aids, the current market also allows for Personal Sound Amplification Products (PSAPs). PSAPs resemble hearing aids in function and appearance; however, PSAPs are not classified as medical devices. These cheaper devices are designed mostly for mildly hard of hearing patients. Several consumer electronics brands such as Samsung, SONY and BOSE have already introduced Bluetooth headsets with PSAP functions. On the other hand, Apple’s AirPod has enhanced its PSAP features via its voice assistant in its iOS 12. They are regarded as the current industry leaders in PSAP and are poised to enter the OTC hearing aid market. RelaJet has already begun to discuss with these companies regarding potential cooperation in OTC hearing aids.

At the same time, RelaJet has continued to be in contact with the world’s six major hearing aid manufacturers. In the near future, consumer electronic device makers and these traditional hearing aid companies will be competing for market share in OTC hearing aids. RelaJet intends to make its multi-talker speech separation a very prominent feature of OTC hearing aids.

Chen has said that besides PSAP and non-prescription hearing aids, multi-talker speech separation can be implemented in many other devices in the future, namely, smart speakers and automotive systems. RelaJet mainly deals with the front-end analysis of speech recognition to achieve “Voice as a Service.” RelaJet’s business model would resemble that of Dolby, which licenses its technologies to consumer electronics manufacturers.

With a Timetable for Product Launch, RelaJet Will Enter CES 2019 to Gain Publicity in the US and the EU

RelaJet, founded in March of 2018, has partnered with an American audio processing chipmaker. Moreover, if manufacturers need to use their technology, they also offer NRE (Nonrecurring engineering, a one-time engineering costs) and other technology licensing services. Presently, RelaJet focuses on the medical device and headphone markets with PSAP features. It is worth noting that the startup is also designing products for a few customers. RelaJet’s short-term goals are to have marketable products containing their technology in the first quarter of 2019 and all OTC hearing aids incorporating their unique voice recognition technology by 2020.

The medium-term goal is to integrate the global audio database. AI algorithms must adapt accordingly to different languages. The six major hearing aid makers have cornered 95% of the market in the US and the EU; therefore, RelaJet will first focus on these two major markets. With the FDA lifting the ban on OTC hearing aids, RelaJet intends to license multiple-talker speech separation to the six major hearing aid manufacturers in 2020.

Lastly, RelaJet intends to attend the upcoming CES 2019 as part of the delegation led by Taiwan Tech Arena. Chen hopes to establish a more strategic partnership with American hearing aid makers and Silicon Valley voice recognition developers.

Facebook Comments