GlowTTS voices for
phoonnx — flow-based acoustic models (text → mel),
paired with a vocoder from
phoonnx-vocoders.
One voice per subfolder (
model.onnx + native
config.json).
58 voices, 15 languages.
Part of the
phoonnx community mirror.
Vocoders across the set: 53 HiFi-GAN · 2 multiband-MelGAN (neural, stats-normalized) · 3 Griffin-Lim.
1from phoonnx.model_manager import TTSModelManager
2m = TTSModelManager(); m.merge_default_voices()
3voice = m.voices["larynx/en-us-ljspeech-glow_tts"].load() # downloads model + linked vocoder
4for chunk in voice.synthesize("Hello from GlowTTS."): ...
Acoustic models are converted to ONNX by a standalone exporter (no coqui-tts /
Larynx package dependency). See
docs/glowtts.md.
