Intelligent voice assistants are on the advance. Smart speakers play music, give weather forecasts, read entire audio books aloud and control devices. Especially those who live in a house with building automation can use voice command to control blinds, lighting, cameras, energy, music and much more.
However, the intelligent voice assistants have one major disadvantage: they listen. Continuously. This is the only way in which the voice assistants can react immediately when a request comes up. As soon as the smart speaker is then activated by a so-called activation word (e.g. "Ok Google"), the speech recognition and processing takes place in the cloud, whereby all data is transmitted over the Internet and stored on the providers' servers. In addition, unintentional activations (e.g. by the television set) or targeted attacks on the devices occur again and again, whereby sensitive data can get into the Internet or even functions in the smart home can be controlled by unauthorized persons from outside.
In this project, we aim at developing a voice assistant that works completely without Internet access and without "listening" microphones. Instead, sensors take over this work by initiating communication with the building user in defined situations. The microphones for voice input are activated for a short time only for the duration of the current dialog. In this way, the system can proactively inquire as to the current needs and plans of the users depending on the situation, and provide appropriate tips and advice.
We are convinced that the future belongs to speech-controlled and intelligent systems. With this project, we want to go a step further and develop a system that can meet the highest standards in terms of security, data protection and privacy.