rhasspy

/

piper-voices

Like 778

Follow

Rhasspy 356

Hello again! This is a second Lithuanian voice, Reginutė's sister:

https://huggingface.co/RobertasTa/lt_LT-ingute-medium

Female, medium, 22,050 Hz, one speaker. She was fine-tuned from lt_LT-reginute1-medium with our own tools, so she uses the same phonemizer and the same phoneme_id_map. Her config differs from reginute1 only in dataset and length_scale (1.15).

Licence: NewGenLTU Open RAIL-D 1.0 (https://sitti.vdu.lt/newgenltu-openrail-d-license/). It is free of charge and does not prohibit commercial use. It does come with use restrictions (Attachment A) that must travel with the voice, so the MODEL_CARD reproduces them verbatim.

Notes for whoever merges this

1. Same dependency as reginute1: "phoneme_type": "lithuanian" (piper1-gpl #296, merged on 2026-09-15). Piper 1.8.0 refuses to load the voice with a clear error, exactly as it does reginute1 today. If you would rather merge this together with the release that carries #296, that is entirely reasonable.
2. Files: the voice folder (.onnx, .onnx.json, MODEL_CARD, samples/speaker_0.mp3) and its voices.json entry, with sizes and md5 from the shipped files. Nothing else in voices.json changes, and _script/voicefest.py passes.
3. Sample: the first line of test_sentences/lt.txt, run through the phonemizer. Please regenerate it the way you do for every voice.
4. Phonemizer parity: the phonemizer in piper1-gpl main and the copy in the voice repository give identical phonemes for this voice (checked on 20,000 dictionary words and the test sentences), so the catalogue version sounds the same as the one in the voice repository.

Thank you for Piper, and for getting Reginutė in so quickly.