October 25, 2018 our-blog 0 Comment

Artificial Intelligence in Acoustic World.

Using artificial intelligence in creating music and art is not very new idea but still an experimental thought. So if you are thinking of working on a project like #magenta, you should know that outcome can be different from your expectation because it required a lot of data to make your artificial intelligence intelligent.

But creating music is an art which require lot of practice practically, so think about it, how artificial intelligence will be able to catch the primitives of music notes?

Then also we human are known to show our interest in doing the things that are next to impossible and although this thought is experimental till now but it has a possibility to get success. But our question will be –why is it important? What will it do? And most importantly how we will be able to do?

WHY

Now a days for a movie or TV series, media industry requires peoples to apply background music but it is financially expensive for the film makers and also it is achieved manually through software which is time consuming too, so we want to solve the problem using AI which doesn’t require dependency to hire other person for doing same task and also it will save time and money.

WHAT

To generate background music for a video with no sound e.g. suppose a horse is riding in the video and you have to compose background music for that scene instantly. In this case what artificial intelligence will do? well, It will capture the scene and model will produce the required music according to the prediction of training data.

HOW

But it is not easy as saying, here we are using Recurrent Neural Network to compose music, hence it requires music in some form of sequence of events unlike mp3 which have low level of visual representation. To compose music it requires music in some form of notations which can be represented it as sequence of events or characters like ABC notations or MIDI.

Challenges

It requires lot of raw music and video data to train the model so that model can predict the outcome music of scene and also it requires a lot of computational power (hardware with GPU). It’s an experimental project, so it’s possible that you may not get as expected result.

Architecture