Artificial intelligence and Machine learning
Before we dive into picking the right method we need to know what AI is.
Artificial intelligence (AI) as the name implies is something created by humans and with the intelligence to understand/think,
it is the study of how to train the computer so that computers can do things which at present time humans can do better.
Machine learning (ML), is a machine that is capable of learning new things without the machine being necessarily programmed to. It’s an use of AI that gives the capability of automatically learning and adapting from past experiences.
The main difference between the two is that AI in general is mainly decision making(finding the optimal solution) and ML is learning new things from given data (go for solutions whether that is optimal or not.)
We want to generate adaptive audio based on the user’s bio metric and computer vision to generate personalized adaptive audio.
This splits the ML into 2 parts: computer vision and audio generation.
Computer vision: For the computer vision part we want to experiment with the kinect, having the kinect pick up movement and tracking the different poses will give us more data to work with.
Personalized audio: Our initial concept with the audio generation was to use the different bio metrics and connect them to different parts of the instruments to generate music,
For example Connecting the Heart with the BPM of the music, using the movement/poses of the users to generate visuals,eye tracking to move the visuals, using neural waves to influence the rhythms and change style of music based on the most preferred beat.
I started with the audio part, there are different neural networks that can be used some are better at certain task than others. I started testing with RNN (Recurrent neural network). this network is able to take input at different kind of lengths, we can either give it audio that's 1
0 seconds long or 10 minutes to learn from.
In this diagram at the bottom we have the input, the hidden layer at he middle and at the top the output.
Here are two examples of what I could generate with RNN using a drum kit midi.
Comments