"AI-Powered Vocal and Accompaniment Separation in Songs"

2019-11-06

For music producers and anyone looking to use isolated audio for performances, splitting songs into separate vocals and accompaniment has always been a headache. There are currently many methods to perform this operation, but the process is time-consuming and the results often have many flaws. A new open-source AI tool makes this challenging task faster and easier.


The software is called Spleeter and is developed by music service company Deezer. Yesterday, the company released the code as an open-source package and made it available for download and use on Github. Simply provide an audio file to Spleeter, and it will divide the file into two, four, or five separate tracks called stems. When running on a dedicated GPU, it can split audio files into four stems 100 times faster than real-time speed.





Technology expert Andy Baio wrote a blog post about Spleeter, which includes many examples. Baio stated that the isolated vocals produced by the software sometimes have a robotic auto-tune feel, but compared to other solutions, the effect is already very realistic."



The software is indeed powerful, but it requires some technical knowledge to use. Unless you regularly use software like Python or Google's AI toolkit TensorFlow (used to train Spleeter), you must download some programs to launch and run Spleeter. Additionally, it must be operated through command-line input rather than a more accessible visual interface.


Deezer pointed out that this is not the first time machine learning has been used to automate such tasks, and the company has already laid the foundation with many early research. Aurelien Herault, Chief Data and Research Officer at Deezer, said that the company trained its software with various types of pre-separated vocals on 20,000 music tracks. With this information, the software learned how to isolate the tracks themselves.



Herault said: Deezer itself uses Spleeter for a range of research applications to help improve its streaming service. Internally, we use it as a preprocessing tool to perform complex research tasks such as music classification, transcription, and language detection.


On the other hand, Spleeter also successfully demonstrates how AI tools make complex creative work easier. Machine learning is currently being used to automatically perform a range of time-consuming tasks, from removing backgrounds in images to upgrading textures in old video games. These tools are increasingly being integrated into software, from Adobe's Photoshop to new fields like Runway ML.


Deezer stated that they have no plans to turn Spleeter into a paid tool, but others can accomplish the same work. Spleeter is the perfect choice for music producers who want to integrate isolated vocals into their mixes or for those who want to create homemade karaoke accompaniment tracks.


Github code link: https://github.com/deezer/spleeter

RECENT AI NEWS

RECENT AI TOOLS