A lip-reading application from Ireland’s startup Liopa is stated to be representative of a revolutionary finding in the domain of visual speech recognition (VSR), which undertakes training of artificial intelligences to read lips with no audio feedback. 

Liopa’s offering, SRAVI (Speech Recognition App for the Voice Impaired) is a communication facilitator for speech-impaired patients. It is probable to be the first lip-reading application put out for customer purchase, going by accounts from Vice/Motherboard. 

Scientists led by a broad array of prospective commercial applications which includes surveillance tools have been carrying out work for several years to train computer devices to read lips, and it has served as somewhat of a challenge. Liopa is undergoing the process to get SRAVI certified as a Class I Medical Device in Europe, with hopes to attain the certification by August. That would enable them to start the sale to healthcare providers. 

Several technological heavyweights are also researching lip-reading artificial intelligences. Scientists connected with or carrying out work directly for Huawei, Samsung, Google, and Sony are all undertaking research on VSR systems and seem to experiencing rapid progression, according to the Motherboard account. 

Liopa gets the second contract for UK Defence and Security Research 

How lip-reading artificial intelligences are being produced and how it might be leveraged are becoming reasons that stimulate concern and even anxiety. Liopa lately made the announcement that it as been chosen to be included in Phase 2 of the DASA Behavioural Analytics initiative, intended at assisting the UK’s Defence and Security Accelerator generate capability in behavioural analytics. These have been defined as “context-specific insights” obtained from information on persons and groups, which could facilitate “dependable forecasting with regards to how they are probable to act in the future,” 

The tool that has experienced much demand would facilitate law enforcement entities to seep through silent CCTV footage and detect when persons utter specific keywords. 

The Liopa VSR engine shoots video of a person(s) talking as input, and leverages artificial intelligence to forecast the subject’s most probable utterances, going by a press release from Liopa, which has its basis in Belfast, Northern Ireland. The engine can be leveraged to detect key words uttered in surveillance video content (CCTV) where audio is likely not available or of bad quality. 

DASA Delivery Manager, Eleanor Humphrey, said, “Behavioural Analytics is a thrilling and emergent capacity that is identifying innovative ways to keep the populace safe from malicious threats. We are delighted to be working with Liopa to accelerate their technologies and look forward to seeing the outcomes” 

Liam McQuillan, Founder and Chief Executive Officer of Liopa, specified in the release, “This contract enables us to build on the progress had in the Phase 1 project. It’s brilliant validation of our VSR technology in a practical scenario that will give invaluable data for Defence and security personnel.” 

Liopa is not isolated in its objective to leverage artificial intelligence for lip-reading functionalities. Surveillance entity Motorola Solutions possesses a patent for a lip-reading framework developed to assist police. Skylark Labs, a startup whose progenitor has connections to the US Defence Advanced Research Projects Agency (DARPA), specified to Motherboard that its lip-reading framework is presently leveraged in private-homes and a state controlled power organization in India to identify foul, unparliamentary, abusive language. 

VSR technologies could be caught up in ethical matters much like face recognition 

A few see the sticky wicket ahead much like what has faced the face recognition market, which has been caught up in ethical issues.  

“This is one of those regions, from my viewpoint, which is a good instance of ‘just because we can do it, doesn’t mean we ought to.’, stated Fraser Sampson, the United Kingdom’s biometrics and surveillance camera commissioner, to Motherboard. “My principal worry in this sphere would not necessarily be what the technology could achieve and what it fell short in, it would be the chilling impact of persons believing it could do what it states. If that then served as a barrier for them to communicate in public, then we wind up in a much bigger domain than merely privacy, and privacy is large enough as it is.” 

Scientists within artificial intelligence are now aware of the ethical connotations of how artificial intelligence is leveraged. For instance, the NeurIPS conference ow needs artificial intelligence scientists to submit, combined with their proposals, impact statements about how their discoveries might influence society at large. 

Stavros Petridis, who has executed connected research at Imperial College London and is now employed by Facebook, talked to Motherboard with regards to the issue. “In the previous year, there have been various discussions in the published literature surround ethical considerations for VSR technologies,” he specified. “Provided that there are no commercial applications available yet, there are very good odds that this time, ethical considerations will be considered prior to comprehensive commercialization of this technology.” 

Liopa Chief Executive Officer Liam McQuillan also stated to Motherboard regarding the matter, stating the organization is at least a year away from having a framework than can go about lip-reading keywords from mute CCTV footage at needed level of precision. He stated that the organization has thought about the potential of a privacy backlash. “There may be concerns here that actually forbid the ultimate leveraging this technology,” McQuillan specified. 

In January, at the Consumer Electronics Show, Sony furnished a bird’s eye perspective view of its visual speech enablement offering which was work in progress, that leverages camera sensors and artificial intelligence for augmented lip reading. Mark Hanson, Sony’s VP of Product Technology and Innovation, specified that the product isolates a user’s lips and converts their movements into words, independent of any noise that is occurring in the background or foreground, going by an account in PCMag.  

The innovative product’s technology only captures lips, and not faces, so no personally identifiable data about the user is retained with the app, so at that level privacy is ensured. Whether the product rises up to meet the regulatory and privacy related challenges on a holistic level, remains to be seen.