PrismVoice-TW
The Taiwan Mandarin voice of Prism Live (Wave Foundry), as the app downloads it.
| File | What | Origin / license |
|---|---|---|
primetts_v21_3voice.onnx |
PrimeTTS v2.1 — MB-iSTFT-VITS, 16 kHz, three Taiwan Mandarin speakers (0 Xinran ♀, 1 Anchen ♂, 2 Bowen ♂). Unchanged. | Luigi/PrimeTTS, Apache-2.0 |
lexicon.json |
Character and phrase readings in bopomofo with a tone digit 1–5, used in place of g2pW on the phone. Built from McBopomofo's BPMFBase.txt, BPMFMappings.txt, heterophony*.list and phrase.occ, with word-level corrections learned from g2pW's readings of 4,000 Taiwan Mandarin sentences. |
McBopomofo data, MIT (OpenVanilla / Mengjuei Hsieh et al.); compiled by Wave Foundry |
cmudict.dict |
English pronunciations (ARPAbet with stress). Unchanged. | CMU Pronouncing Dictionary, BSD |
Inputs of the ONNX model, as in PrimeTTS: x, tone, lang (int64, [1, T], a blank id 0 between symbols), x_lengths, sid, noise_scale (0.667), length_scale (1.0); output wav (float, 16 kHz). The symbol table is PrimeTTS's frontend_bopomofo.py order.