Japan's SoftBank Develops AI Filter That Turns Angry Screams Into Calm Speech

Have you ever been screamed at for no fault of yours? Are you employed in a job where you have to bear the brunt of hearing your clients shout all the time? Call centre employees have it hard, having to deal with angry clients most of the time, which can affect their mental peace. However, artificial intelligence (AI) has a remedy for almost everything these days and this is no exception.

A Japanese multinational company has developed an artificial intelligence filter that can detect angry screams and translate them into calm speech to reduce the stress experienced by call centre operators.

Large tech companies have recognised the daily stress experienced by their call centre operators, who often bear the brunt of customers’ anger and frustrations. To address this, some companies have implemented stress management programmes featuring relaxation techniques, meditation, yoga, or therapy to alleviate irritability and anxiety. However, a Japanese company named SoftBank might have found a more effective solution by using AI to remove screaming and aggressive speech from the interactions. SoftBank asserts that they have spent three years developing a voice filter capable of detecting and converting screams into calm speech.

Toshiyuki Nakatani, one of the developers at SoftBank, said, “We have developed an emotion suppression system that will protect the employees from the angry voices of customers that distress them.”

SoftBank’s voice filter works through a two-step process. Initially, the AI detects an angry voice and identifies key speech elements. In the next step, it employs acoustic tools to convert the speech into a more natural and polite tone. Importantly, the filter does not modify the words spoken, but it softens the intonation significantly.

This means that call centre operators will still hear any insults, but in a gentler tone, which is intended to help reduce their stress and anxiety. To train the AI, SoftBank’s engineers had 10 actors record at least 100 common phrases, including screams, accusations, threats and demands for apologies. In total, over 10,000 voice recordings were used to develop the AI filter.