Skip to content

Voices

20 voices, 10 languages. Every profile is enrolled by this project’s own pipeline from a recording made or released for speech-technology use: personal donations recorded for TTS, and CC0 / CC-BY corpora whose terms allow it. The donor or source, the licence and a sample are named for every voice. The full chain lives in docs/voices/roster/provenance.json: consent basis, reference construction, SHA-256 of profile, reference and sample, and seed.

Listen to all twenty voices and compare each generated sample with its enrollment reference.

Profiles ship on the Hugging Face repository under voices/, versioned next to the checkpoint they enrol against.

The reference SHA-256 identifies the original WAV used for enrollment. Those source WAVs are not redistributed in the model repository; reference.public_preview names the Opus derivative played on the demo page, not the bytes that were enrolled.

We have evaluated English by ear. We do not speak the other nine languages well enough to judge their naturalness reliably. Feedback from native speakers is very welcome.

voicelanguagegendersourcelicence
joeEnglishMOHF-Voice donationsCC0
kathleenEnglishFOHF-Voice donationsCC0
carmenSpanishFCML-TTSCC-BY-4.0
daveSpanishMOHF-Voice donationsCC0
coletteFrenchFKyutai tts-voicesCC-BY-4.0
henriFrenchMKyutai tts-voicesCC-BY-4.0
kerstinGermanFOHF-Voice donationsCC0
thorstenGermanMThorsten-VoiceCC0
danteItalianMMLSCC-BY-4.0
paolaItalianFOHF-Voice donationsCC0
darkmanPolishMOHF-Voice donationsCC0
gosiaPolishFOHF-Voice donationsCC0
inesPortuguese (European)FCML-TTSCC-BY-4.0
tugaoPortuguese (European)MOHF-Voice donationsCC0
nathalieDutchFOHF-Voice donationsCC0
pimDutchMOHF-Voice donationsCC0
nilsSwedishMNST SwedishCC0
selmaSwedishFNST SwedishCC0
frejaDanishFNST DanishCC0
sorenDanishMNST DanishCC0

Ten seconds of clean audio is enough:

import loudkit as lk
mine = lk.enroll("my-recording.wav", "loudreader/loudr-1", name="my-voice")
mine.save("voices/my-voice.safetensors")

Consent is yours to obtain. See RESPONSIBLE_USE.md.