The world now has technology like Siri that helps check the weather for tomorrow. Another similar AI assistant is Alexa to plays favourite songs using speech recognition technology. However, these technologies necessitate the use of audible speech. What if a person cannot talk or if vocalized speech is inappropriate in a certain situation?
Thus, researchers have unveiled a device named Speechin. It refers to a silent-speech recognition (SSR) device that uses images of skin deformation in the neck and face. The information is collected by a neck-mounted infrared (IR) camera to detect silent orders. The technology is a massive contribution to World Elderly and Disabled Assistive devices Market. It would help disabled people and the elderly use AI assistants to carry out normal tasks. Further, humans can also use it generally in situations where speaking would be inappropriate.
There are two issues here. First, why a necklace? Second, why is speech silent?. The team believes that a necklace is a more familiar form factor than ear-mounted gadgets, which may not be as pleasant. As far as silent speech, individuals may assume they already have a speech recognition gadget on their phones. But one needs to vocalize sound for them, which may not always be socially suitable, or the person may not have the ablility to vocalize speech.
The team added even with silent speaking; this technology has the ability to learn a person's speech patterns.
Researchers involved had created the original prototype in 2020. They stated that now they are introducing an altogether new form factor, new hardware, into this field.
The device looks a lot like NeckFace, a technology the team showed off last year. The necklace uses infrared cameras to collect photos of the chin and face from beneath the neck. It can then track entire facial expressions in real-time.
Speechin, a necklace, has an infrared camera set on a 3D-printed necklace casing that is hung on a silver chain and points up at the wearer's chin. The developers added a wing to each side and a coin to the center for further stability.
A necklace-mounted IR camera could be preferable to a standard camera in front of the face for two reasons. A camera in front of an individual's face is photographing what's behind you, which presents privacy concerns.
The researchers recruited ten people to quietly recite a list of 72 one-syllable "nonwords" made up of phonemes (combinations of 18 consonants and four vowels) to push the system's limits. As per the tests, the team concluded that SpeeChin detects commands with an average accuracy of 90.5 percent in English and 91.6 percent in Mandarin, respectively.