What it does
MYELIN decomposes a track into four domains, spatial, spectral, dynamic and compound, and maps trigger events across its running time. The output is a temporal map of where the nervous system is likely to respond and which signal property is responsible.
One documented analysis run over a 237 second track surfaced 190 multi-modal triggers and 638 onset surprises, with a peak frisson score of 0.85. Those numbers are the instrument reporting on a specific track, not a benchmark.
The measurement it is really making
Frisson is mediated by the dopaminergic reward system and the sympathetic nervous system. The caudate activates in anticipation, before the emotional peak; the nucleus accumbens activates during the release after it. Roughly half of people experience frisson at all, and those who do show different white matter connectivity between auditory cortex and emotional processing regions.
The published literature on this is correlational. No study has successfully manipulated musical stimuli to support a causal claim. MYELIN moves toward predictive detection from the audio signal alone, with no listener in a scanner and no skin conductance hardware in the loop.
In the code
@dataclass(frozen=True)
class Trigger:
t: float # seconds into the track
domain: str # spatial | spectral | dynamic | compound
strength: float # 0..1, normalized within the track
cause: str # the signal property responsible
# A compound trigger is one where two or more domains fire inside the
# same window. Those are the ones that correlate with a reported
# response, so they are counted separately rather than summed in.Illustrative shape. Keeping the cause on the trigger is what makes the output a map rather than a score: a peak you cannot attribute to a signal property is not a finding.
How this differs from the ordinary version
It is a coherence instrument that happens to take audio
CORTEX makes manufacturing coherence visible. EPPE makes civic coherence visible. CLOS makes cognitive coherence visible. MYELIN makes emotional coherence visible. Same architecture, different substrate. Reading it as a music tool understates what it is for.
Prediction rather than correlation
Detecting the response after the fact from a body sensor is the established method and it requires a subject and hardware. Predicting it from the waveform is a different claim, and it is falsifiable against exactly the biological signal the lab already streams.
In the field
The closed loop that does not exist yet
A cardiac device streaming heart rate variability during playback would supply ground truth labels for the detector: biological response as signal, as label, as training data, produced by one operator who also makes the music. Every component is in place. The loop itself is not built, and it is listed as an open question rather than a result.
Questions
- Does MYELIN need a listener wired to sensors?
- No. The detection runs on the audio signal. Biological data is the intended validation path, not an input requirement.
- What are the four domains?
- Spatial, spectral, dynamic, and compound markers that combine the first three.