{"id":"styletts-2","nom":"StyleTTS 2","editeur":"Columbia University","pays":null,"poids_ouverts":true,"poids_selon":null,"licence":null,"licence_selon":null,"liens":{"page":"https://github.com/yl4579/StyleTTS2"},"fiche_auto":true,"page":"https://quelleia.com/voix/modeles/styletts-2/","releve":"2026-10-19","en_bref":"92e sur 96 en synthèse vocale en anglais (Artificial Analysis), dans un peloton de 6 modèles que la marge d'erreur ne permet pas de séparer.","presentation":"Modèle de synthèse vocale qui génère le style de la voix par diffusion, sans extrait de référence, et qui s'entraîne avec de grands modèles de langage vocaux (comme WavLM) servant de discriminateurs. Les auteurs proposent des modèles préentraînés sur LJSpeech (une seule voix) et sur LibriTTS (plusieurs voix, adaptation à un nouveau locuteur en zéro-shot).","sortie":"2023-06-13","langues":{"annoncees":{"annoncees":"anglais pour les modèles préentraînés (corpus LJSpeech et LibriTTS)"},"prises_en_charge_selon_open_tts":null,"evaluees":["en"]},"preuves":[{"appuie":["presentation"],"lu":"https://raw.githubusercontent.com/yl4579/StyleTTS2/main/README.md (« StyleTTS 2 differs from its predecessor by modeling styles as a latent random variable through diffusion models to generate the most suitable style for the text without requiring reference speech »), 17/10/2026"},{"appuie":["presentation"],"lu":"https://raw.githubusercontent.com/yl4579/StyleTTS2/main/README.md (« Furthermore, we employ large pre-trained SLMs, such as WavLM, as discriminators with our novel differentiable duration modeling for end-to-end training »), 17/10/2026"},{"appuie":["presentation"],"lu":"https://raw.githubusercontent.com/yl4579/StyleTTS2/main/README.md (« Moreover, when trained on the LibriTTS dataset, our model outperforms previous publicly available models for zero-shot speaker adaptation. »), 17/10/2026"},{"appuie":["langues","presentation"],"lu":"https://raw.githubusercontent.com/yl4579/StyleTTS2/main/README.md (« The pretrained StyleTTS 2 on LJSpeech corpus in 24 kHz can be downloaded at »), 17/10/2026"},{"appuie":["langues","presentation"],"lu":"https://raw.githubusercontent.com/yl4579/StyleTTS2/main/README.md (« The pretrained StyleTTS 2 model on LibriTTS can be downloaded at »), 17/10/2026"},{"appuie":["langues"],"lu":"https://raw.githubusercontent.com/yl4579/StyleTTS2/main/README.md (« It contains the pre-trained [PL-BERT](https://arxiv.org/abs/2301.08810) model, which was pre-trained on English (Wikipedia) corpus only. »), 17/10/2026"},{"appuie":["sortie"],"lu":"https://arxiv.org/abs/2306.07691v1 (« [Submitted on 13 Jun 2023 (this version), latest version 20 Nov 2023 ( v2 )] »), 17/10/2026"}],"classements":[{"id":"synthese/artificial-analysis","nom":"Synthèse vocale en anglais, Artificial Analysis","rang":92,"sur":96,"valeur":891.73,"unite":"Elo","bas":876.73,"haut":906.73,"votes":2584,"chiffres":{"prix_par_million_caracteres":2.82,"date_sortie":"2023-06-13"},"voisins":["google-neural2","google-standard"],"source":{"nom":"Artificial Analysis","mention":"Source : Artificial Analysis","url":"https://artificialanalysis.ai/text-to-speech/model-families/styletts","releve":"2026-10-19"}}],"noms_dans_les_sources":{"artificial_analysis":"StyleTTS 2"}}