pub fn phenomenal_voice_frequency_hz(t: &Tensor10D) -> f32
Frequency + FM for worklet — uses phenomenal σ mapping (not bin peak alone).