‘Creative jobs are immune to AI invasion’ remains the old chestnut in most ‘future of jobs’ discussions. However, the advancements in technology, specifically AI, is proving otherwise.
Wayne Cheng, Founder and Machine Learning Engineer at AUDOIR, an AI-assisted music creation software provider, recently explained how advanced AI technology can be used to create music.
“Today’s most advanced AI technology is deep learning, an artificial neural network that maps between data domains,” Wayne said.
Deep learning
Within this artificial neural network, a chain of networks maps data from one dataset to another (say, A and B). The AI learns relevant features of A that can map into B during the mapping or Automatic Feature Extraction phase.
The deep learning AI phases consist of two parts: Training, and Prediction. At the training phase, the AI is given two datasets, and it has to learn how to map from one dataset to another (A to B). In the Prediction phase, the AI is given new data (A) and is tasked to predict the new dataset (B).
Songwriting AI most commonly leverages two deep learning technologies — Autoencoder architecture and Autoregressive models. That is, music notes can be predicted using past sequences of music notes. At present, the most popular autoaggressive AI architecture includes RNN, RNN with attention mechanism and Transformer.
Once the AI Is trained, it can generate a large number of derivative music. However, the process does not just end there. Songwriting AI should have a plagiarism checker and a critic baked in to check the veracity and the quality of the music.
“Keep in mind that the quality of the training data will determine the quality of the generated music. It is essential to find the highest quality data when training the AI,” Wayne said.
Amazon DeepComposer, Google Magnate, OpenAI MuseNet, AIVA, and Amper Music are the major players in songwriting AI.
Tech behind SAM
According to the leading music streaming platform Spotify, the top one percent of the artists account for 90 percent of the streaming on the platform.
US-based AUDOIR wants to change just that. It aims to provide equal opportunities to all stripes of artists in the music industry, using the power of AI. AUDOIR, with the help of SAM, a suite of AI-powered songwriting tools, helps artists write songs.
SAM’s AI neural engine is trained with thousands of hit songs. Once the training is complete, SAM can create a hundred new songs within seconds. SAM further checks for plagiarism and then ranks the hundreds of songs it creates based on quality to make the selection process easier for its users, listing the top ten songs. The entire process essentially takes about three minutes.
Furthermore, SAM has agents to help in the process of music composition:
Agent R: Creates music from scratch
Agent T: Creates music from lyrics or text
Agent C: Creates music from chord progression
Agent N: Creates music from motif
Agent E: Extends motif
Agent L: Converts text to lyrics
So far, AUDOIR’s SAM has churned out songs across genres including Pop, EDM, R&B, Rock and Country, at free of cost.