AI improves the listening experience

Clear sound, even when you are surrounded by noisy traffic or your colleague is hammering on their keyboard? A team at Fraunhofer IIS developed a low-complexity real-time AI-based noise reduction technology. What makes it unique is that it is device independent.

Edwin Mabande beams with pride about his team’s achievement: within two years, the Communication Acoustics Group at Fraunhofer IIS has developed an AI-based noise reduction algorithm that is specifically tailored for use on mobile devices and which is now part of the Fraunhofer upHear Microphone Processing technology suite. “Noise reduction for mobile devices has been around for many years,” Group Manager Mabande says. Up until now, such technologies were mostly based on traditional signal processing techniques, which use mathematical models. Artificial Intelligence takes noise reduction to the next level and could be the answer to an increased demand for better sound – be it on smartphones, which are often used in noisy environments, or on the telephone and video conferencing platforms that have become an integral part of every workday. “The continuous trend to mobile work has greatly fueled the demand for noise reduction solutions, as many people can’t ensure quiet surroundings for every call,” Mabande explains.

Clear sound that doesn’t drain the battery

The need to develop an AI-based solution for smaller, mobile devices was the main challenge for the Fraunhofer team. “Previous solutions required very powerful processors,” Mabande says. That amount of power is not always available on, say, smartphones, so a far less complex technology was needed. “We had to reduce the complexity significantly to make sure that the mobile device battery wouldn’t be completely discharged within minutes,” Mabande says. At the same time, the technology had to be able to perform noise reduction and signal transmission almost simultaneously. This can best be done by AI-based solutions that use Deep Neural Networks (DNN). “The problem was that most existing DNNs were very general in design and therefore highly complex,” he recalls.

Real-time noise reduction on mobile devices

So the team decided to develop their own solution – and can now look back on a very successful project. Most importantly, their technology is significantly less complex than existing solutions. Mabande is certain that “our AI-based noise reduction is one of the most powerful and efficient systems on the market.” It can also be implemented on a broad range of devices and is independent of the hardware a manufacturer uses, which is something most previously available solutions couldn’t provide. Other advantages of the Fraunhofer noise reduction are its robustness and its outstanding performance in all scenarios tested until now. Such tests deliver important feedback to the developers: After all, a child speaks at a different pitch than an adult, and there are different sounds in Mandarin than in German or English. “The technology always works,” Mabande says. All these benefits make the noise reduction solution a valuable part of the Fraunhofer upHear product range, which comprises audio technologies created by Fraunhofer IIS that enhance sound from recording to playback. “Right now, Artificial Intelligence is a hot topic at all technology companies. With our noise reduction technology, we can take on a leading role,” explains Sebastian Meyer, Product Manager for the upHear solutions. This is also evident from the conversations about licensing options for the software that he currently has with various manufacturers in the audio industry.

Article by Julian Hörndlein, Freelance journalist and PR copywriter