AI in Music Production

 

Artificial Intelligence rather quickly is slowly becoming a mainstream tool for content creators. Recently released OpenAI’s GPT-3 is generating a lot of buzz in the natural language processing community. The package contains a much improved and much larger language model promising compelling text generation capabilities. There is no question that the time when humans communicate with computers using spoken or written English or other languages is rapidly approaching.

Composing music with AI models

The situation in the music production business is similar, although the market forces did not push developers as hard as in the financial or e-commerce sectors of the economy. Markov model-based algorithms are composing music for decades now. Still, the new era of sophisticated music production tolls like DAWs, virtual and hardware synthesizers, effects solutions, creates a rich platform for AI algorithms.

Most of us consume music through streaming applications. We create massive data sets of musical preferences by our clicks, ratings, frequency of play, duration, and many other explicit and implicit feedback mechanisms. For machine learning algorithms, this is a perfect task, i.e., look at the patterns in the audio files, compare those with user feedback datasets and create models that will tell us what kind of music to produce. Just imagine that your Karma-driven algorithmic sequencer in your music workstation communicates with a machine learning model created from millions of user ratings. The model does not have to replace a talented musician in the creative processes, but it will tell you right away how well the song will sell.

Making the song sound great in various circumstances

The creative music-making process is not the only one that will be affected by the AI. In the past, CD players, amplifiers, and big speakers were the way to go when it comes to high-quality listening systems. Today, the majority of music is streamed to multiple devices. Many digital streaming services are lossy, meaning that the quality of sound is radically downgraded to ensure smooth downloads. Also, many of us spend pennies on cheap headphones that tend to break as frequently as the expensive ones.  It means that the tone, volume or dynamics, and other characteristics of records might be very far from those recorded in the studio.

There are dedicated software packages that utilize AI to ensure that radically distorted streamed records still sound excellent and consistent in most circumstances. Typically, this is when the mastering process needs an experienced sound engineer. However, more and more artists decide to accomplish this costly task on their own using an increasing range of intelligent software choices.

https://www.theverge.com/2019/1/30/18201163/ai-mastering-engineers-algorithm-replace-human-music-production

This problem is very similar to the one faced by software engineers that make to ensure that their code executes well on all mobile, desktop, console, and cloud devices. It appears that AI will take over music sooner or later. Perhaps, we will completely automate the tedious parts of building musical records to the point that creativity and imagination will be the only things required. Decades of learning music theory and mastering piano or violin will not be necessary anymore. Sadly, the creative process can be modeled by machine learning quite effectively.