About Us
With the recent advancements in various aspects of Artificial Intelligence, AI-generated music has gained enough prominence and notability throughout the past few decades. Music generation algorithms are constantly evolving and trying to find patterns in which musical notes should be placed to compose original melodies. Hence why UNSUNG.AI is exploring the role of machine learning as a tool in the creative process and investigating AI’s impact on the music industry. The model’s music creation process involves a collaboration between AI and humans to achieve the final artistic vision.
Lyrics Generation with GPT-3 and enCore.ai
As we received an invite to OpenAI’s beta version of the third-generation transformer GPT-3, we utilized its playground to come up with suitable lyrics for our songs. We also trained the encore.ai model on lyric generation.
Beat Generation with Aiva
We used Aiva to generate melodies we which merge, repeated and added effects to create the beats we used for the songs.
Composition and Autotuning with eMastered
We fine tuned the singing produced by the AI model, and a bit of autotune so everything flows properly.
Song Cover Generation with Rokulu
We generated covers for the songs based on the presumed title from the lyrics generated by the AI models.
Reflection on Project
Biggest Challenges Faced in the Project?
The two biggest challenges we faced were training the AI model to generate lyrics and the second was actually finding an AI generated voice to sing the lyrics since most AI voices sounded too robotic.
Insights Gained form the Project?
This project really opened our eyes to how humans and AI could collaborate to actually create a full musical composition. We were responsible for polishing up the singing and beats generated by the AI to create actual music.
What Parts of this Project do you Think Worked Really Well?
The beat generation was the part which worked the best. The AI was excellent at generating pieces of beats which we repeated and combined to create the complete usable beat.
What Parts of this Project did not Work Very Well In Your Opinion?
The AI model we used to sing was by no means perfect. It lacked a proper musical flow and sometimes sounded very robotic.
Key Take-aways from the Project?
There is great potential for collaboration between AI and humans especially in the area of music composition, and this project goes to prove that.