A method of determining from audio information and vehicle conditions whether a vehicle or an occupant is at risk, and triggering an action in response. For instance, microphones may detect the sound of an infant left in a car and detect a dangerous vehicle temperature. The owner of the car may then be messaged and the car alarm may sound. Alternatively it may detect the sound of a jack and the tilt of the vehicle to determine theft of vehicle wheels. A first neural network may determine whether a sound originated from inside or outside the vehicle, and a second neural network may classify the sounds into a plurality of categories. The neural networks may learn by processing audio files with mel-frequency cepstrum coefficients and performing logarithmic and linear Fourier Transform of the audio on a Mel scale. The vehicle conditions may be one or more of; a key not being inserted in the ignition or within the vehicle, the vehicle being lifted or in motion, the doors being locked, and the internal temperature exceeding a threshold. Also disclosed is an audio recognition system with first and second neural networks, the system processing audio files with mel-frequency cepstrum coefficients.
Vehicle ambient audio classification via neural network machine learning
2018-06-06
Patent
Elektronische Ressource
Englisch
Vehicle ambient audio classification via neural network machine learning
Europäisches Patentamt | 2019
|Vehicle ambient audio classification via neural network machine learning
Europäisches Patentamt | 2021
|Vehicle Ambient Audio Classification Via Neural Network Machine Learning
Europäisches Patentamt | 2019
|Vehicle Ambient Audio Classification Via Neural Network Machine Learning
Europäisches Patentamt | 2018
|