Data Science

AI Helps Discover Hit Songs for Record Labels

How does a top of the charts song sound? This new AI system can tell you. Musiio, a new music tech company based in Singapore, is using deep learning to identify the best songs for record labels and streaming services. The service can listen to over 30,000 songs a day, around the same number of new songs released every day.
The startup was founded by Hazel Savage, a music-tech expert with 12 years of experience in the music industry, and Aron Pettersson, a machine learning developer who’s been coding for 17 years.
Musiio scans for multiple traits of a track including recording quality, the perceived skill of the artist, a probability of success, comparable percentile, and more, the team said.
“Musiio won’t replace the need to have people listening to music,” Hazel Savage, the company’s founder told TechCrunch. “But we can delete the inefficiencies.”
Using NVIDIA GeForce GTX 1080 and GeForce GTX 1080 Ti GPUs with the cuDNN-accelerated TensorFlow deep learning framework, the team trained their neural network on thousands of hits and non-hits. The team trains their system on audio rather than stats and data from third-parties.

An example of how the Musiio system works.

The startup is also using the same GPUs used for training for inference.
“By ‘listening’ to more tracks than a human could ever comprehend and identifying characteristics and patterns therein, our AI allows you to better predict success,” the company said. “We know that by solving the volume issue in music, and by offering actionable insights, you can increase your ‘hit-rate’ and ultimately your revenue,” they stated.

The team is currently working with record labels and streaming services. They are also offering their services to the Free Music Archive, a Creative Commons-like site developed by independent U.S. radio station WFMU.
The company is also in the process of speaking to investors to raise a seed round of funding.
Read more >

Discuss (0)