Hello @RaphaelVilani.
The “combinatorial MIDI follower” (see the file midi_follower.asco.txt attached in the thread) does not require a MIDI note-off event. You only need to signal the onset of the expected MIDI note at the appropriate time, for example by setting setvar $pitch_in.
If you use the MIDI mode of the listening machine, then yes, a note-off event is required, for exactly the reason you mention.
If you are interested in using another listening machine, you may want to look at a paper that will appear at ICMC 2026:Automatic Hybrid Following in Real-Time Mixed Music: A Case Study with Antescofo and ipt˜ for Flute Playing Techniques (1.0 MB). The paper explains how several listening machines can be used together and how one can switch between them when needed.
José Miguel Fernandez has developed two approaches based on gesture following. In Sources rayonnantes , he uses the Gesture Follower developed by the ISSM team to track the gestures of the conductor, who is equipped with RiOT sensors. The Gesture Follower performs a gesture-to-gesture alignment with a prerecorded reference gesture. Anchor points are defined on this reference gesture; when the live gesture reaches one of these points, a notification is sent to Antescofo (using setvar). This notification triggers (via whenever) a next_event, allowing the system to advance in the score.
The score itself remains a standard Antescofo score, where electronic actions are anchored to musical events. In this configuration, however, the musical events are not detected by the built-in listening machine but are instead notified by the Gesture Follower through the setvar / whenever / next_event mechanism.
This type of setup—essentially following the conductor—has also been used in other works, such as Idea by Sampo Haapamäki, see also IDEA project: Conducting Gestures Dataset | Ircam Forum. If you are interested in pursuing this direction, you should contact Serge Lemouton, who has implemented several such setups.
In the longer-term project Gekipe, developed by José Miguel Fernandez together with HEM Genève, a Kinect sensor tracks the position of the performer’s hands relative to the torso. The hands are also equipped with RiOT, and the accelerometer data streams are sent to Antescofo together with the hand position data. Some preliminary processing of these data streams is used to categorize gesture types (for example kicks, smooth directional motions, etc.). The detected gestures and their positions are then interpreted contextually within the score. This approach is therefore performer-centered, and if you want to explore it further, you should contact José Miguel Fernandez.
Finally, a remark regarding your last statement:
“For that to happen, Antescofo should follow the conductor as it does with instruments, both for tempo inference and for dealing with mistakes by the conductor or by the beat-recognition system…”
It is important to note that Antescofo has no concept of musical measures. Musical events are specified without bar information: the score is represented as a flat sequence of elementary musical events (notes, chords, trills, etc.), rather than as measures containing beats.
Moreover, when using the Gesture Follower to track a conductor, the system does not really consider the notion of a “mistake.” Instead, it continuously aligns the incoming accelerometer data with the prerecorded gesture, stretching or compressing the time axis when necessary. In this framework, gestures are not “missed” or “skipped”; rather, the system simply assigns different probabilities to the gestures stored in its gesture dictionary during the alignment process.
Hope this helps.
there are different ways to achieve the same think.