According to Japan's The Asahi Shimbun, Softbank started a study on "emotion cancelling" three years ago. This technique uses AI voice-processing technology to alter a person's voice during a phone call.

SoftBank employee Toshiyuki Nakatani had the idea after viewing a TV programme about customer harassment.

"If the customers' yelling sounded like Kitaro's Eyeball Dad, it would be less scary," he said.

To enhance vocal conversions, the voice-altering AI mastered numerous expressions, including yelling and accusatory tones. Ten actors were engaged to perform over 100 phrases with various emotions, training the AI with more than 10,000 pieces of voice data.

The technology does not alter the wording, but the pitch and inflexion of the voice are softened.

 For example, a woman's high-pitched voice is lowered in tone to sound less resonant. A man's bass tone, which might be intimidating, is raised to a higher pitch to sound softer.

However, if an operator cannot discern if a customer is angry, the operator may not be able to respond appropriately, which could just upset the customer further. Hence, the developers ensured that a slight element of anger remains audible.

According to the company, the most significant burdens on operators are hearing abusive language and being ensnared in lengthy conversations with customers who will not disconnect -- such as when making persistent requests for apologies.

With the new technology, if the AI determines that the conversation is excessively long or too abusive, a warning message will be dispatched, such as, "We regret to inform you that we will terminate our service."

The company intends to further enhance the technology's accuracy by having AI learn voice data and hopes to market the technology starting in fiscal 2025.

Nakatani said: "AI is adept at handling complaints and can do so for extended hours, but angry customers desire a human to apologise to them. AI will become a mental shield that prevents operators from overstraining their nerves."