Automatic Music Transcription

Now that the Audiophile’s Analyzer is complete how can all of the features be provided to the Musician’s Workbench?

Simply backporting the code would not be the best solution, it would result in massive code duplication. Even more problematic would be the clash of design philosophies.

The Musician’s Workbench was designed to reproduce the functionality of the original SA-10 hardware. It is lean and real-time by design.

The Audiophile’s Analyzer was designed to provide every known music transcription technique, and to provide it all in a single integrated package. It is large and not real time.

The solution is to allow the Audiophile’s Analyzer to import session files from the Musician’s Workbench.

Here we see the Beats graph from the Analysis tab of the Audiophile’s Analyzer. The beats are from the session file, and were originally generated by the metronome of the Musicians workbench. The audio signal is overlaid, this is the original audio sampled by the Musician’s Workbench. The audio was only sampled on the beat when there was sound, and only enough to allow a single FFT to be performed.

The spectrogram for the same audio is shown.

Now the user can re-transcribe the session using any of the techniques available in the Audiophile’s Analyzer including the built in CNNs. This would simply not be possible in real time as it requires 7 FFTs to be performed before the spectrogram can be sliced and sent to the CNN.

The Audiophile’s Analyzer can also provide some insight into the internals of the Musician’s Workbench which are not normally displayed. Answering questions such as, “Should I have used Delayed Sampling?”, “Did I select the best Octave Range?”.