File size: 2,058 Bytes
ae20b77 0b4486f bb696ac 0b4486f 968ad34 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 | ---
license: cc-by-4.0
language:
- it
tags:
- text-to-speech
- tts
- piper
- italian
datasets:
- committa/serena-synthetic-it-27h
pipeline_tag: text-to-speech
---
# it_IT-serena-medium — Piper voice (Italian)
Custom Italian voice for [Piper](https://github.com/rhasspy/piper), trained from the
`committa/serena-synthetic-it-27h` dataset (Qwen3-TTS synthetic voice).
- Language: Italian (`it`) — neutral Standard Italian (no regional accent)
- Voice: **female** ("serena")
- Sample rate: 22050 Hz
- Quality: medium
- Files: `it_IT-serena-medium.onnx` + `it_IT-serena-medium.onnx.json` (must stay together)
## License
CC-BY-4.0 — this model is derived from
[committa/serena-synthetic-it-27h](https://huggingface.co/datasets/committa/serena-synthetic-it-27h).
## Recommended inference settings (config A)
The `.onnx.json` already embeds these defaults; the CLI uses them automatically.
They were chosen after stability testing: the Piper stock defaults (0.667 / 0.8)
tend to swallow phonemes (e.g. /r/ in "portato").
| Parameter | Value |
|-----------------|-------|
| noise_scale | 0.6 |
| length_scale | 1.0 |
| noise_w | 0.4 |
| sentence_silence| 0.2 |
## CLI usage
```sh
piper -m it_IT-serena-medium.onnx < text.txt
```
Explicit overrides (optional):
```sh
piper -m it_IT-serena-medium.onnx --noise-scale 0.6 --noise-w 0.4 < text.txt
```
## Home Assistant (piper add-on)
The Home Assistant add-on ignores the `.onnx.json` settings and always passes its
own options, whose defaults (noise_scale 0.667, noise_w 0.333) are NOT optimal for
this voice. For the best quality set in the add-on options:
- `noise_scale`: **0.6**
- `length_scale`: 1.0
- `noise_w`: **0.4**
- `sentence_silence`: 0.2 (optional; add-on default is 0.0)
## Notes
- Pronunciation of rare vocabulary (<10 occurrences in training data) can be off;
fix with espeak-ng rules if needed.
- The training checkpoint (base for finetuning other Italian voices) is published at:
https://huggingface.co/committa/it_IT-serena-medium-checkpoint
|