Space voice order: best to weakest (male_voxcpm_2, female_voxcpm_2, male_voxcpm_1, female_voxcpm_1); default male_voxcpm_2

Defaults for every voice: temperature 0.8, tempo 0.8 (voices_index, voices.json, card); samples of the new voices regenerated at these defaults

female_voxcpm_1: 90 Hz high-pass post-filter (removes p pops learned from its VoxCPM2 reference clip): highpass_hz in voices_index/voices.json, note in the card, filtered samples

Three more synthetic voices (female_voxcpm_1, male_voxcpm_2, female_voxcpm_2): one Matcha model + own HiFi-GAN each, onnx/<voice>/, voices_index.json; card and attribution for four voices

Questions: keep '?' glued (as in training); neutral style 8

Default temperature 1.0

Voice renamed fav2 -> male_voxcpm_1; question-mark tip; browser cleaner keeps 'word ?'

Matcha-TTS-PL v2: fav2 voice distilled from VoxCPM2, English words, fav2 vocoder