YAML Metadata Warning: empty or missing yaml metadata in repo card
Check out the documentation for more information.
Alma — Swedish Female Voice for Piper TTS
Overview
Alma is a Swedish (sv_SE) female voice for the Piper text-to-speech system. Named after a real person, the voice was trained to provide natural-sounding Swedish speech synthesis for accessibility applications, smart home systems, and assistive technology.
Model Details
| Property | Value |
|---|---|
| Language | Swedish (sv_SE) |
| Gender | Female |
| Quality | Medium |
| Sample Rate | 22050 Hz |
| Model Size | ~60 MB (ONNX) |
| Phonemizer | espeak-ng (sv) |
| Architecture | VITS (via piper-train) |
| Training Epochs | 8000 |
| Training Data | NST Swedish TTS Dataset |
Usage
With Piper CLI
echo "Hej! Jag heter Alma." | piper \
--model sv_SE-alma-medium.onnx \
--config sv_SE-alma-medium.onnx.json \
--output_file output.wav
With Home Assistant
Place the .onnx and .onnx.json files in your Piper voices directory and restart the Piper add-on.
With NVDA (Sonata add-on)
The model is compatible with the Sonata NVDA add-on for screen reading.
Training Data
Trained on the NST Swedish TTS Dataset (Språkbanken, National Library of Norway). The dataset contains high-quality studio recordings of Swedish speech.
License
The model weights are released under CC BY 4.0, following the license of the NST training data.
Author
Daniel Nylander (@yeager)
Performance
- Real-time factor: ~0.04x on RTX 4080 (25x faster than real-time)
- Suitable for real-time applications on modern hardware
- Compatible with Raspberry Pi via Piper's optimized ONNX runtime
Samples
Audio samples are included in this repository (sample1.wav, sample2.wav).