Hear the sound to avoid danger? This AI can accurately defend against theft and traffic accidents.

category:Internet
 Hear the sound to avoid danger? This AI can accurately defend against theft and traffic accidents.


Chris Mitchell, chief executive of Audio Analysis, said: This is equivalent to a language model, but it only applies to sound.

Audio Analysiss voice recognition AI system AuditoryNET has a wide range of applications. The company is working with Bragi, a German earphone maker, to produce headphones with AI support that prioritize user safety by identifying ambulance alarms and alerting users of their approach.

Mitchell said the idea was to protect users from dangerous situations because people are making themselves totally deaf in cities.

Audio Analysis also embedded Auditory NET in smart home devices such as Hive Home Hub360, which detects smoke and carbon monoxide detectors, barking dogs or broken windows. AuditoryNET runs without Internet connections, which is good news for privacy-conscious people.

To train AuditoryNET algorithms, Audio Analysis uses data from Alexandria, the worlds largest audio data collection platform. What weve been doing recently is mapping our entire voice world, Mitchell said.

Figure 2: Audio Analysiss Voice Map contains 6 million audio files, including six types of restaurant window glass cracking sound.

Audio Analysis has created a sound map containing more than 6 million audio files. It looks like a rainbow hive: it visually represents a color block of similar sounds gathered together, or tone island.

Mitchell said that by adjusting the volume, their technology can recognize slightly different sounds. For example, there are six kinds of breaking sound from restaurant glass. The breaking sound of laminates made of glass is quite different from other sounds.

Alexandrias audio sample data pool must be built from scratch. Common sounds in movies are useless because they are not real. If you go to see the latest Avenger League movie, Im sure the Hulk will drop something on the car, Mitchell said.

That car alarm doesnt make a real car alarm, he said. This voice will train AI to face false realities, often catastrophic, such as the need to detect whether someone has broken into a car. So the company set up its own lab to capture the original sound and sent teams to record other sounds in various environments. They even rely on volunteer networks to enter sound labs or use recorders to capture voices in the real world. Audio Analysiss future plans are still under wraps, but its goal is clear: to continue to develop technologies to recognize voices inside and outside the home, including those that sound like alarms. (Selected from: WIRED Author: Will Bedingfield Compiler: Netease Intelligent Participation: Small) Source of this article: Netease Intelligent Responsibility Editor: Gu Yuxin_NBJS8596

That car alarm doesnt make a real car alarm, he said. This voice will train AI to face false realities, often catastrophic, such as the need to detect whether someone has broken into a car. So the company set up its own lab to capture the original sound and sent teams to record other sounds in various environments.

They even rely on volunteer networks to enter sound labs or use recorders to capture voices in the real world.

Audio Analysiss future plans are still under wraps, but its goal is clear: to continue to develop technologies to recognize voices inside and outside the home, including those that sound like alarms. (Selected from: WIRED Author: Will Bedingfield Compiler: Netease Intelligent Participation: Small)