Pluggable window classifier.
The default implementation is signal-processing based and dependency-free. A neural classifier (e.g. YAMNet) can be dropped in behind this interface without touching the detector state machine or the session hub.
Readonly
Stable identifier reported in logs and calibration output.
Scores one analysis window.
Mono samples in [-1, 1], oldest first.
[-1, 1]
Sample rate of samples in Hz.
samples
Music/speech likelihoods plus the underlying features.
Pluggable window classifier.
The default implementation is signal-processing based and dependency-free. A neural classifier (e.g. YAMNet) can be dropped in behind this interface without touching the detector state machine or the session hub.