Markov-chain music generation builds a probability table of “what note is most likely to follow what,” based on a training melody. Then it samples from that table to create new sequences in the same style. The application is controlled from a DAW using MIDI and has a build in sequencer which responds to an external midi clock. To use the application a song or multiple songs can be created. The song requires training data to work. After this this training data can be used to generate new musical phrases based upon recorded input.

1. Collect MIDI training data To generate a solo, the markov model requires a substantial amount of training data to capture your playing style. We recommend recording solos without chords in Ableton Live and exporting them as a single MIDI file. You can also use the trainingsdata midi file (included in the assets folder and performed by Jeroen Ermens) to get started right away. For best results, provide a MIDI file containing around one hour of solo material for a specific song, though effective results can already be achieved with as little as 15 minutes of MIDI data.

2. Install and run the application Follow the steps in Requirements, Installation and Running the applicaiton described above.

3. Integrate MIDI messaging with Ableton Live (MacOS) To use Signal Inference in a live setting, you can connect it to your existing Ableton Live set via MIDI. The first step is to ensure that Ableton is properly configured to communicate with Signal Inference over MIDI. Follow these steps:

   1. On your Mac navigate to Audio MIDI Setup. Click on your IAC driver, and create 3 ports named: CLOCK, TO_DAW, FROM_DAW. Click on Apply and go back to Ableton Live
   2. In Ableton Live go to Settings > Link, Tempo, Midi and select the following boxes:
      • For Midi Inputs, select for your [INPUT PORT]: Track, Sync, Remote, MPE
      • For Midi Output, select for for TO_DAW: Track, Sync and select for [OUTPUT PORT]: Track
   3. Go into arrangement view and insert a new MIDI track and name it MIDI_TO_SIGNAL_INFERENCE. This MIDI track will be used to send all commands to the application. Instead of "No Ouput", select IAC (FROM DAW) as output channel
   4. In the Clip View of this channel, you can add envelopes (you don't have to add notes!) by clicking on Envelopes.

Signal Inference uses the following MIDI CC messages to generate solos. This cheat sheet will guide you through integrating them step by step into your Ableton Live set by assigning the CC values to envelopes.

CC message: 32 -> Start recording
Set this message at the beginning of any solo you play and record if you want to use it as training data for the markov model to generate from. 

CC message: 33 -> Stop recording
Set this message at the end of your solo—or at the start of the markoved chain-generated solo—to stop the recording.

CC message: 40 -> Generate solo
Set this message immediately after CC message 33, at the point where you want the generated solo to begin generating. The value of this CC message is the amount of bars that you want to generate. This value is passed on to the application. 

CC message: 48 -> Start playback
Set this message immediately after CC messages 33 and 40, at the point where you want the generated solo to begin playback of the generated solo.

CC message: 49 -> Stop playback
Not required, but can be used if you want to stop the playback before the amount of bars that is played is not finished yet.

CC message: 56 -> Swith song. 
Set this CC message at the start of a song. This ensures the track will switch to the right song. This value of this message correspongs to the MIDI Selection message that you set for each song in the Songs tab of the applicaiton. For example: Song "Greem Unions" has a MIDI selection message of 1, then the MIDI CC control message of this song should be set at the start of the this song and should have a value of 1. 

Dragging an envelope up or down changes the value of a CC message. For certain CC messages, the value itself has a specific meaning (for example, the number of bars to generate or the song being selected). If no specific meaning is indicated, setting the value to 127 (to send the message once) and then returning it to 0 is sufficient. Using this technique, you can create a short “block” signal to trigger the message.

Apply the CC messages at every point where you want Signal Inference to generate solos. Once connected, you’ll see the incoming MIDI messages reflected in the interface: the Play, Generate, and Record buttons will turn blue when triggered, and the song selection in Signal Inference will automatically update (highlighted in blue) when another song is played in Ableton.

Are you working with a Ableton Live set from scratch? We recommend you to make a separate click track with the correct BPM for each song, and backingtrack to give context to the solo's that are being generated. Check out the example how that is done.

Check out the example Ableton Live set (in /assets/example). You can use this in combination with the c (also in /assets/example) to see how the MIDI connections are set up.

4. Play with Signal Inference: Once you’ve set up all your MIDI commands in Ableton, most of Signal Inference’s core functionality will be ready to use. From there, you can refine the output by tweaking parameters, experimenting with different training datasets, and applying timing corrections. When you’re happy with the results, save the model for future use