Songscription launches a “shazam for the scores” powered by AI “

A small business called Song Launched last week with AI models that automate musical transcription, transforming an audio file of a song into scores in a few minutes. Operating on a freemium model, the product is aimed at professional and amateur musicians.

“We hope to make music more pleasant,” said Songscription and student of the MBA / MA program in Stanford. “We imagine a future where a high school teacher from Nebraska Rural (will be) capable of obtaining partitions for the songs that their students want to play, (and) that said music will be organized specifically for the group instruments and offered at the individual level of play of each student.”

At launch, Songscription can transcribe music for several different instruments, although the piano model is the most reliable. In the future, the company hopes to add different transcription outings (for example, guitar tabs), as well as arrangements for a full band as opposed to a single instrument.

This type of product could be useful for a musician who records a song on which they work, then download it to get the score – in this way, they can ignore the step to manually transcribe their work. And for those who cannot read and write scores, Songscription will also generate a piano roll, which shows a digital representation of music played on a virtual piano.

Users can also automate musical transcription directly from YouTube links. Downloading a file forces users to check a box to confirm that they have the right to transcribe the file, but it would be easy to simply check the box and obtain free partitions for songs protected by copyright.

“For music learners … Since you are allowed to listen to a song, note the notes by ear and interpret it on your piano at home (as long as you do not charge any performance), it is not entirely clear that the use of a technical platform to give you a step in advance crosses a legal border, although we understand that the domain evolves and that our application can enter a gray area.”

Legality around a large part of the way we engage with these creative IA tools is to be debated, however Recent judicial decisions seem to promote technological companies compared to artists. However, Songscription does not create new music generated by AI -AI – it provides a tool for musicians to speed up the manufacturing process of their own tabs or guitar partitions.

“Since our platform allows users to modify scores, we position ourselves as a increased musical notation software that helps people accelerate the transcription process,” said Carlins.

The underlying architecture of the Songscription AI model is based on a paper This co-founder Tim Beyer published, alongside researcher Angela Dai.

In order to obtain the training data necessary to create this type of AI model, Songscription works with certain musicians who were ready to share or sell their piano performance and their partitions. The company has also used partitions in the public domain, although the majority of training data are synthetic, said Carlins. In this case, Songscription would convert the partitions to audio, then modify the files to simulate the conditions of the real world with a background noise or a reverberation.

Only seven months after its foundation, Songscription collected money before the Reach Capital seeds and will participate in Stanford Startx Accelerator.