rhasspy

/

piper-voices

like 632

Follow

Rhasspy 311

New Italian voice for the catalog: serena — female, neutral Standard
Italian (no regional accent).

• Language: it_IT (Italian, Italy)
• Quality: medium, 22,050 Hz
• Dataset: committa/serena-synthetic-it-27h (CC-BY-4.0)
• Training: trained from scratch (94 epochs, ~27 h of audio)
• Recommended inference settings (embedded in the .onnx.json):
  noise_scale 0.6 / length_scale 1.0 / noise_w 0.4

Original model repos:

• Voice: https://huggingface.co/committa/it_IT-serena-medium
• Training checkpoint (base for finetuning other Italian voices):
  https://huggingface.co/committa/it_IT-serena-medium-checkpoint

Files follow the piper-voices layout and pass the voicefest.py checks
(structure, config, naming). voices.json was regenerated accordingly.

New Italian voices for the catalog: serena - female, neutral Standard
Italian (no regional accent), in two quality tiers.

• Language: it_IT (Italian, Italy)
• Quality: medium and high, 22,050 Hz
• Dataset: committa/serena-synthetic-it-27h (CC-BY-4.0)
• Training: trained from scratch (medium: 94 epochs, high: 99 epochs, ~27 h of audio)
• Recommended inference settings (embedded in the .onnx.json):
  noise_scale 0.6 / length_scale 1.0 / noise_w 0.4

Original model repos:

• Voice (medium): https://huggingface.co/committa/it_IT-serena-medium
• Voice (high): https://huggingface.co/committa/it_IT-serena-high
• Training checkpoints (base for finetuning other Italian voices):
  medium: https://huggingface.co/committa/it_IT-serena-medium-checkpoint
  high: https://huggingface.co/committa/it_IT-serena-high-checkpoint

Files follow the piper-voices layout and pass the voicefest.py checks
(structure, config, naming). voices.json was regenerated accordingly
(entries for both it_IT-serena-medium and it_IT-serena-high).