MODUS TTS — Fallout 76 Voice Model
Piper TTS voice trained on MODUS dialogue from Fallout 76.
Models
| File | Epochs | Training samples |
|---|---|---|
modus_10000.onnx |
10,000 | 421 voice lines |
modus_10000_v2.onnx |
10,000 | 682 voice lines |
Both share the same base, settings and hardware. The voice character is near-identical; v2 is noticeably more robust and slurs less, especially on words outside the training vocabulary. Use v2.
Specs
| Property | Value |
|---|---|
| Base checkpoint | en_US-lessac-high |
| Quality | high |
| Sample rate | 22,050 Hz |
| Format | ONNX |
| Language | English |
| Batch size | 8 |
| Precision | 16-bit AMP |
| GPU | NVIDIA RTX A2000 12GB |
| Training time | ~4–5 days per run |
Training vocabulary: 1,984 unique words across 11,117 tokens. Median line length 17 words.
Usage
pip install piper-tts
wget https://huggingface.co/petrusilius/modus-tts/resolve/main/modus_10000_v2.onnx
wget https://huggingface.co/petrusilius/modus-tts/resolve/main/modus_10000_v2.onnx.json
Both files must sit in the same folder.
echo "We have you now, General." | \
piper --model modus_10000_v2.onnx --output_file output.wav
Useful flags:
| Flag | Effect |
|---|---|
--length_scale 1.3 |
slower |
--length_scale 0.8 |
faster |
--sentence_silence 0.5 |
longer pause between sentences (default 0.2) |
Input notes
Plain text only, no SSML.
- Write numbers as words:
forty two, not42 - Spell out abbreviations:
General, notGen. - Use
...for mid-sentence pauses - Keep sentences short for better prosody
Limitations
- Mispronounces words absent from the training data. This is a vocabulary limit, not an epoch limit — more training does not fix it. Keep generated text close to the model's own register, or supply IPA via espeak-ng.
- Long unpunctuated sentences sound rushed.
- VITS samples at inference, so output varies slightly between runs on identical input.
Disclaimer
Non-commercial fan project. Fallout 76 and all related assets are property of Bethesda Softworks.
- Downloads last month
- 77