Softbank plans to cancel out angry customer voices using AI.

Japanese telecommunications giant SoftBank recently announced that it has been developing "emotion-canceling" technology powered by AI that will alter the voices of angry customers to sound calmer during phone calls with customer service representatives. The project aims to reduce the psychological burden on operators suffering from harassment and has been in development for three years. Softbank plans to launch it by March 2026, but the idea is receiving mixed reactions online.

According to a report from the Japanese news site The Asahi Shimbun, SoftBank's project relies on an AI model to alter the tone and pitch of a customer's voice in real-time during a phone call. SoftBank's developers, led by employee Toshiyuki Nakatani, trained the system using a dataset of over 10,000 voice samples, which were performed by 10 Japanese actors expressing more than 100 phrases with various emotions, including yelling and accusatory tones.

Voice cloning and synthesis technology has made massive strides in the past three years. We've previously covered technology from Microsoft that can clone a voice with a three-second audio sample and audio-processing technology from Adobe that cleans up audio by re-synthesizing a person's voice, so SoftBank's technology is well within the realm of plausibility.

By analyzing the voice samples, SoftBank's AI model has reportedly learned to recognize and modify the vocal characteristics associated with anger and hostility. When a customer speaks to a call center operator, the model processes the incoming audio and adjusts the pitch and inflection of the customer's voice to make it sound calmer and less threatening.

For example, a high-pitched, resonant voice may be lowered in tone, while a deep male voice may be raised to a higher pitch. The technology reportedly does not alter the content or wording of the customer's speech, and it retains a slight element of audible anger to ensure that the operator can still gauge the customer's emotional state. The AI model also monitors the length and content of the conversation, sending a warning message if it determines that the interaction is too long or abusive.

The tech has been developed through SoftBank's in-house program called "SoftBank Innoventure" in conjunction with The Institute for AI and Beyond, which is a joint AI research institute established by The University of Tokyo.

Source: https://arstechnica.com

Share: