Classify Speech Emotion (Dimensional)

How it fits

Typical backends

Caveats

• State lifetimeThe model and the device are read once at startup, so changing either takes effect only after a container restart.
• CompatibilityAudio is resampled to 16 kHz internally regardless of the upstream sample rate, matching the model's training rate.
• AccuracyOutput is continuous regression over the valence-arousal-dominance space, not a discrete emotion label; the three values are each roughly bounded and have no calibrated probability interpretation.
• I/O contractEmpty audio yields a neutral midpoint reading rather than failing, so downstream consumers always receive a complete triple.
• LimitationA single affect estimate is produced for the whole frame, so very long segments average over their internal variation; segment by speech turn upstream when finer granularity matters.