Project news #1

Posted on

It’s time to give a quick overview about what I did lately and what I plan to do. Though everything move forward slowly there is some work achieved. I thought about many projects to do as well but it will come at a later time.

On the side of the Acoustic Scene Classification, The work on the preprocessing has been done and it’s computing as we speak (calculate frame by frame the spectrograms). I will soon implement the neural net using probably Tensorflow or maybe PyTorch, but most of all I will have to create an instance on Amazon AWS in order to use GPU and speed up the process if I don’t want to die of old age before seeing some results. The ultimate goal will be at some point to use this work for the Ears Wide Open project that is stalled for several months now.

I continued to work as well on porting my draft thesis on auditory motion perception from LaTeX to Sphinx (python). Many things to do on the side of the images but almost all the chapter on the Filehne illusion is done. At some point, I’ll probably try to port the experiment on a python framework and create a head tracker using cheap sensors. I will maybe try to continue the work on the bibliographic side in order to keep up with the recent development on this subject and I wanted to read more about the different point of views regarding the fundamental aspects of sound motion perception.

I recently wanted to go back to Pure Data. I discovered that Pd-Extended has been stopped and a replacement got birth under the name of Pd-L2Ork. By installing that I saw that my library was a real pain in the ass to install and some abstrations that I wrote would benefit a lot from being rewritten in C in a more generic way. By saying that I’m talking about the different abstractions that were dedidacted to sound motion over 24 speakers. I would be better and interesting to have objects where you can specify the number of channels.

I added few functions on the [audioquest library][l5]. Nothing fancy but always useful such as sine, sine sweep signals, conversion of ERB to Hertz and vice versa. This library is really for study and not intended for production but it’s a fun project to work on from time to time.

I have other ideas that I keep track of that would more about hacking. I will probably talk about it later on when some work will have been made.