Effective separation of sound musicaccompaniment to various channels is essential for music producers, music lovers, performers, creators of karaoke applications or producers. Separation of vocals from percussion instruments, pianos or other musical instruments is a rather painstaking task requiring a professional approach and sophisticated equipment.
Specialists associated with the streaming serviceDeezer announced a learning neural network with an efficient algorithm for dividing sound sources into channels. Open source is the open source Spleeter application, created on the basis of Python and TensorFlow and using a machine learning technique.
For all users who downloadapplication and library, the following functions will be available. The division into two channels: vocals and any musical accompaniment. The division into 4 sound channels: vocals and drums, bass and other music. And the division into 5 channels of vocals, drums, bass, piano and other audio accompaniment.
The first users have already noted that the applicationvery quickly splits the audio file into the necessary channels. For example, working with a 5.5 minute audio file takes only three minutes to process. At the same time, a high quality of the ongoing separation into sound channels was noted, a similar one to which not one of the previous techniques has yet been able to achieve.
Unlike pre-existing applications that were paid (for example, $ 4 for processing one track), the presented Spleeter program will not charge a fee.