phoonnx community mirror
Collection
phoonnx mirrors of community TTS — VITS, MMS, Matcha, OptiSpeech, GlowTTS + vocoders — each with a clean phoonnx config.json for TTSVoice.load. • 11 items • Updated
phoonnx-vocoders
Neural vocoders (mel → waveform) for two-stage phoonnx voices
(GlowTTS, Matcha-TTS, OptiSpeech). One vocoder per subfolder; each ships a
vocoder.json declaring its vocoder_type and any mel-preprocessing flags
(e.g. stats_norm with mel_mean/mel_std). Part of the phoonnx community mirror.
| subfolder | type | source |
|---|---|---|
wavenext-22khz |
wavenext (raw) | BSC-LT/wavenext-mel (apache-2.0) |
alvocat-vocos-22khz |
vocos (ISTFT) | projecte-aina/alvocat-vocos-22khz (cc-by-nc-4.0) |
vocos-mel-22khz-univ |
vocos (universal) | BSC-LT/vocos-mel-22khz |
larynx-hifigan-vctk-small |
hifigan | rhasspy/larynx (MIT) |
tr-common-voice-hifigan |
hifigan | coqui-ai/TTS |
be-common-voice-hifigan |
hifigan | coqui-ai/TTS |
en-ljspeech-multiband-melgan |
melgan (stats-norm) | coqui-ai/TTS |
uk-mai-multiband-melgan |
melgan (stats-norm) | coqui-ai/TTS |
A vocoder only works with an acoustic model whose mel features match it (sample rate, FFT/hop, n_mels, fmin/fmax, normalization). See the vocoder guide.
alvocat-vocos = Vocos finetuned on Catalan (Matxa); vocos-mel-22khz-univ = universal Vocos for any HiFi-GAN 80-mel (Mixer-TTS etc.). Different mel domains — not interchangeable.