renders a voice through an OfflineAudioContext (no speakers), draws its STFT spectrogram (log-freq) + waveform, and reports metrics. device-free, decisive. use ?voice=NAME (or ?scene=marshNight).