Accurate Music Transcription Software
AnthemScore is a program designed to automatically create sheet music from audio (MP3, WAV, etc.) as accurately as possible. Notes are identified with a state-of-the-art convolutional neural network trained on over 2 million data samples.
The output is a MusicXML file you can view and edit with any standard music notation software. If you choose to install MuseScore, a popular free notation program, sheet music will open automatically when processing is finished. In addition to transcription, you can also:
- Slow down the tempo
- Display the frequency spectrum of the left and right audio channels separately
- Click to play notes and visualize expected harmonics
- Zoom in and listen to short sections
- Overlay note lines, switch the time and frequency axes, and customize the display
To learn more about harmonics and spectrograms, jump to the quick guide.
If you have experience with note recognition programs, you probably know that transcribing music is challenging and accuracy tends to be very low. AnthemScore is able to produce more accurate results using a convolutional neural network trained on polyphonic music. You can read more about how it works here. Currently, no instrument information is extracted, so notes from all instruments get combined into one part.
The downside is that it's currently a little slow. It usually takes at least a couple minutes of processing per minute of audio. This type of note recognition network is still relatively new, so stay tuned for further updates.
You can download and try it out on Windows, OS X, and Linux.
Below are some examples of music automatically transcribed with AnthemScore. The accuracy depends on the complexity of the song and the instruments used. You can generally expect the best results from piano pieces.