Demo

Listen

Twenty voices across ten languages, one seeded sample each. Filter by language, then press a name.

All 20EnglishSpanishFrenchGermanItalianPolishDutchPortuguese (European)SwedishDanish

• joeEnglish · male
• kathleenEnglish · female
• daveSpanish · male
• carmenSpanish · female
• henriFrench · male
• coletteFrench · female
• thorstenGerman · male
• kerstinGerman · female
• danteItalian · male
• paolaItalian · female
• darkmanPolish · male
• gosiaPolish · female
• pimDutch · male
• nathalieDutch · female
• tugaoPortuguese (European) · male
• inesPortuguese (European) · female
• nilsSwedish · male
• selmaSwedish · female
• sorenDanish · male
• frejaDanish · female

Each sample is one seeded render of a public-domain sentence. The source, the licence and the hashes of every voice are in Voices.

From a recording to speech

Each voice below is enrolled from one short recording. Press the left clip to hear that recording. Press the right clip to hear the same speaker read a sentence the recording never contained.

Both clips are files in this repository. Nothing is generated in your browser. The reference is the audio the voice profile was enrolled from, at 48 kHz mono. The generated clip is one seeded render on the reference path, so the same text, voice and seed are repeatable on the same build and backend. Different backends are checked against the project’s measured parity bands.

joeEnglish · male

Reference, 11.0 sthe enrollment input

Generated, seed 11from the profile enrolled above

Alice was beginning to get very tired of sitting by her sister on the bank, and of having nothing to do: once or twice she had peeped into the book her sister was reading.

Alice's Adventures in Wonderland (Lewis Carroll)

0.924 speaker similarity: the cosine similarity between the speaker embedding of the reference and the speaker embedding of the generated clip.

daveSpanish · male

Reference, 11.0 sthe enrollment input

Generated, seed 7from the profile enrolled above

Subió una mona a un nogal, y cogiendo una nuez verde, en la cáscara la muerde, con que le supo muy mal.

La mona (Felix Maria de Samaniego)

henriFrench · male

Reference, 6.6 sthe enrollment input

Generated, seed 7from the profile enrolled above

Maître Corbeau, sur un arbre perché, tenait en son bec un fromage. Maître Renard, par l'odeur alléché, lui tint à peu près ce langage: Hé! bonjour, Monsieur du Corbeau.

Le Corbeau et le Renard (Jean de La Fontaine)

kerstinGerman · female

Reference, 9.8 sthe enrollment input

Generated, seed 7from the profile enrolled above

Ach, was muss man oft von bösen Kindern hören oder lesen! Wie zum Beispiel hier von diesen, welche Max und Moritz hießen.

Max und Moritz (Wilhelm Busch)

paolaItalian · female

Reference, 11.0 sthe enrollment input

Generated, seed 7from the profile enrolled above

C'era una volta... Un re! diranno subito i miei piccoli lettori. No, ragazzi, avete sbagliato: c'era una volta un pezzo di legno.

Le avventure di Pinocchio (Carlo Collodi)

inesPortuguese (European) · female

Reference, 10.4 sthe enrollment input

Generated, seed 7from the profile enrolled above

O meu amigo Jacinto nasceu num palácio, com cento e nove contos de renda em terras de semeadura, de vinhedo, de cortiça e de olival.

A Cidade e as Serras (Eca de Queiros)

0.960 speaker similarity: the cosine similarity between the speaker embedding of the reference and the speaker embedding of the generated clip.

Where a card shows a speaker similarity, that number is measured for the exact two clips the card plays. It is not measured for every voice, so the line is absent where there is no measurement. The figures, the sources, the licences and the hashes are in Voices.

Voice cloning

Ten seconds of clean audio is enough. A cloned voice is a file of tensors around 150 KB, not a model.

import loudkit as lk

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

The weights this needs are published, so the snippet runs today. The method, the audio requirements and the measured similarity are in Cloning a voice. Consent is yours to obtain: Responsible use.