Automatic sonification of video sequences through object detection and physical modelling
Student thesis: Master Thesis and HD Thesis
- Andrea Corcuera Marruffo
4. Term, Sound and Music Computing (Master Programme)
The sounds generated by the objects that surround us is intrinsic to our life. We
associate some specific sounds to the objects, its characteristics, and the actions that
generate them, and we expect to hear the corresponding sound when we see that
item. Similarly, one expect to see the corresponding object when we hear its sound.
For example, on the street, when we hear a characteristic sound of a motor, we know
that a car is approaching. There is, therefore, a particular relationship between the
objects and the sound that they produce.
In films, many of these sound effects are added in post-production, a method
called "Foley". In this project, these sound effects will be generated automatically
based mainly on one characteristic of the objects involved: their material. A system
based on an object detector, an impact detector and a sound modeler will be presented.
A perceptual evaluation in which the subjects will watch some videos and
listen to the sounds predicted by the model will be performed. In this way, it will be
tested if the sounds proposed by the model are played in accordance of what they
expect to hear.
associate some specific sounds to the objects, its characteristics, and the actions that
generate them, and we expect to hear the corresponding sound when we see that
item. Similarly, one expect to see the corresponding object when we hear its sound.
For example, on the street, when we hear a characteristic sound of a motor, we know
that a car is approaching. There is, therefore, a particular relationship between the
objects and the sound that they produce.
In films, many of these sound effects are added in post-production, a method
called "Foley". In this project, these sound effects will be generated automatically
based mainly on one characteristic of the objects involved: their material. A system
based on an object detector, an impact detector and a sound modeler will be presented.
A perceptual evaluation in which the subjects will watch some videos and
listen to the sounds predicted by the model will be performed. In this way, it will be
tested if the sounds proposed by the model are played in accordance of what they
expect to hear.
Language | English |
---|---|
Publication date | Jun 2017 |