AI Generative Drum Transcriptions: A Comparative Analysis

I’ve been waiting for this moment for months! Artificial intelligence has gained ground in various aspects of our lives since its inception. I use AI every day to maximize and fasten my workflow. Over the past months, I have researched how to apply this technology to drum transcriptions. In this article, I review and analyze two websites that can automatically transcribe and generate the drum sheet music of any song. For the following comparison, I chose three of my drum transcriptions: “Aja” by Steely Dan, “Rosanna” by Toto, and “Harridan” by Porcupine Tree.

Drumscrib

The first website I came across was Drumscrib. It’s possible to upload audio files or search for the desired songs to transcribe, and the software will automatically generate the drum sheet music and the synthetic drum track. The transcriptions are generated and provided with essential info (duration, bpm, time signature, note count) and are available for download in PDF format or as mp3, MuseScore, and MIDI files. Prices are accessible, starting from €3 for one drum sheet music + synthetic drum track.

Upon reviewing the transcriptions, the website works well for beginner and intermediate-level songs despite some errors. For advanced songs, transcriptions are inaccurate. For example, on “Harridan” the software transcribed the song in 4/4, while it’s 5/4. The final result is weird and requires effort to edit the file yourself. However, the website is still in beta, and as stated, improvements are coming.
In summary, Drumscrib looks promising and has the potential to become an excellent tool for transcribing drum parts in the future.

  • Fast service.
  • Accessible prices.
  • Acceptable notation.
  • Transcriptions are not accurate.
  • Missing details: dynamics, accents, song section markers.
  • It doesn’t work well on odd-time signatures.

Klangio – Drum2Notes

The second website is Drum2Notes by Klangio. The company provides AI software tools for transcribing music to notes for various musical instruments. The service is free for the first 20 seconds of transcription, and many features are available with paid monthly and annual plans. It’s possible to upload audio files or search for the desired songs. In addition, it’s possible to specify some details such as the time signature, tempo, note duration quantization, and triplets recognition.

The transcription process is fast, but the final result is totally inaccurate. I tried with “Harridan” the first time, and the transcription generated was so weird I had to repeat the process because I thought it was piano sheet music… but it was drum sheet music for real.
I assure you that I’ll give the website another chance in the future; for now, it’s ineffective!

  • Fast service.
  • Transcriptions are totally inaccurate.
  • Weird notation.
  • Missing details: dynamics, accents, song section markers.

To conclude, even if inaccurate, the use of generative AI for transcribing drum tracks seems promising. Drummers should never forget transcribing by ear is an essential skill for a real musician. Artificial intelligence is just an additional/optional tool and not the only way to approach the art of transcribing.
By the way, here are my transcriptions of “Aja,” “Rosanna,” and “Harridan,” available for purchase in my store. I spent an average of 6 hours transcribing note-for-note, editing, and revising these transcriptions. You can easily compare them with the generated sheet music and make your considerations; I’d be happy to hear your thoughts!


Leave a comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.