In recent years, the use of techniques based on machine learning has increased and expanded to different areas. The increasing computing capacity and the availability of public datasets have allowed the development of these techniques, particularly deep learning, that today are considered the state of the art in topics such as pattern recognition, natural language processing,  gaming, among others. In this work, we propose the use of machine learning techniques for the synthesis of vehicle noise. Audio and video data of passing-by vehicles in non-controlled environments are collected in order to build a vehicle pass-by dataset. Automatic dataset annotation and synchronization is performed using a deep neural network, for vehicle identification, and tracking algorithms. Also, geometry projection algorithms are used to determine the vehicle position and average speed. Vehicle noise emission directivity is extracted from microphone array recordings, using beamforming techniques and will be used to train a neural network for audio synthesis. Different neural networks architectures, such CNNs, RNNs, Autoencoders and GANs, will be considered when designing the synthesis network.  Video recording examples and preliminary results about the extraction of vehicle characteristics such category, position and speed will be presented.

Melden Sie sich hier an um die Einladung zu dem Vortrag per E-Mail zu erhalten.
Register here to receive the invitation to this talk via E-Mail.

The lectures are currently offered as a in-person event and as an online stream via the Zoom platform. There are 25 seats available to attend in the lecture hall. To do so you must identify yourself as vaccinated, recovered or tested against SARS-CoV-2.

Zoom-Meeting: 910 0030 1035,
Password: 980455