dhammagift/piper-ru-puriso

Puriso: a Russian male voice for Piper (medium, 22 050 Hz)

Puriso is a Russian male voice for Piper. Its name is Pali for "man". It was made for Dhamma.Gift, where it reads Russian translations of the Pali Canon. Piper has only a few Russian voices, and we hope this one is useful beyond our project.

How it was made

• Base: fine-tuned from Piper's ru_RU-ruslan-medium.
• Teacher: trained on Russian speech synthesized by Supertone's Supertonic voice M1. The input text carried our own stress marks, so the stresses are taught, not guessed. In other words, Puriso is a distillation of M1 into a small, fast Piper model that runs on a CPU, a phone or in the browser.
• Phonemes: espeak-ng ru, as in any Piper Russian voice. Dhamma.Gift adds its own fixes for Pali names in Cyrillic and for a few stresses (dhammagift/pali-tts, respell.py). The voice works without them too.

Listen

• samples/dn22-tak-ya-slyshal.mp3: «Так я слышал. Однажды Благословенный проживал в стране Куру…»
• samples/dn22-sozercanie.mp3: «Вот, монахи, монах пребывает в созерцании тела как тела…»

Use

pip install piper-tts
python -m piper -m ru_puriso2-medium.onnx -f out.wav -- "Так я слышал."

Licence

Both of the following apply:

• CC BY-NC-SA 4.0, from the base voice's dataset (the RUSLAN corpus): attribution, non-commercial, share-alike.
• BigScience OpenRAIL-M: as a model trained on a Supertonic model's output, Puriso is a Derivative of the Model under that licence. Its use-based restrictions (Attachment A, in LICENSE-OpenRAIL-M) apply to you and to anything you build from Puriso. Changes from the original: a new voice distilled from M1's output, see above.

Made by Dhamma.Gift. Attribution: "Puriso voice by Dhamma.Gift, based on Piper ruslan and Supertonic M1".