v0.9.1

Removing code used during research in the production branch

Ensemble method

The ensemble method is now working End-To-End.

Running the algorithm has changed but the original HMM C++ program can still be run the same way.

v0.7.0

• Support for midi
• Support for custom key-profiles and custom key-transitions

v0.6.0

Minor bug corrections. After running the algorithm in all the different key-profile combinations, the best combination has been selected as the default.

This code has been submitted to MIREX.

Main program working and evaluation mode enabled

The full workflow of the program has been implemented. The program can now take an input audio file and output the main key hypothesis. An evaluation mode has also been added to the program, if the key is provided in the name of the input file, the program scores the hypothesis of the algorithm using the same criteria used by the MIREX Audio Key Detection task.

v0.4.0

The main change of this version is a brand new Viterbi algorithm implemented from scratch that is much faster than the previous one. Minor changes to the HiddenMarkovModel class and the way KeySequences are requested once they have been computed.

Second layer HMM enabled

The workflow from a pitch-class sequence to computing the main key is complete.

HiddenMarkovModel is working

Currently, the code can take a sequence of pitch-classes and compute the key from them.

Everything from the point where the sequence of pitch-classes is known has been implemented and it is working.

There is code for an independent program that can make use of the chromagrams in the nnls-chroma plugin and compute the chromagram. The next step in the project is to make all of this a single workflow that takes audio and outputs a key.