Facebook is working on an artificial intelligence that it hopes could one day detect people’s emotions based on their tone of their voice, aiming to alleviate the frustrations of modern voice speaker systems such as Alexa.
Engineers at the social network’s research labs are working out how to train its voice-controlled video chat device, Portal, to understand when a user is angry, an employee said during a tech conference in San Francisco. The system could one day be used across Facebook Messenger and WhatsApp calls, but could lead to privacy fears about the scope of the company’s data collection. Initially, the company is working on using the technology to improve the Portal device, which is controlled by voice and lets users make video calls to friends and family.
If the machine detects hallmarks of anger and frustration in a person’s voice, such as talking loudly and slowly, it might respond differently to commands.For example, instead of offering impatient users a number of options when it is told to “call John”, it will take its best guess at the right action to speed up the call. It is unclear when, or if the feature will be released.
Facebook engineer Parthath Shah admitted while discussing the technology during a ReWork summit in San Francisco that “we still don’t have the answer yet” and that “there is no perfect way to measure satisfaction”. To try and get as much voice data as possible, the company is using employees and paid volunteers to re-enact conversations to work out teething problems. A Facebook spokesperson said: “Our teams explore new technologies that may or may not ever be released as part of a product.” Facebook currently oversees the communications of 2bn people every single day, yet has fallen far behind its rivals when it comes to voice control. It has yet to launch a voice assistant to rival Amazon’s Alexa or Apple’s Siri but has made strides in the past year after installing a “conversation” team led by Rushin Shah, who left his post as head of natural language processing for Apple in 2017. —Courtesy The Telegraph
