{"id":"boson-ai-higgs-tts-3-4b","nom":"Higgs TTS 3","editeur":"Boson AI","pays":"États-Unis","poids_ouverts":true,"poids_selon":null,"licence":"Boson Higgs TTS 3 Research and Non-Commercial License","licence_selon":null,"liens":{"page":"https://www.boson.ai/blog/higgs-tts-3","poids":"https://huggingface.co/bosonai/higgs-tts-3-4b","essayer":"https://docs.boson.ai/models/higgs-tts/overview"},"fiche_auto":false,"page":"https://quelleia.com/voix/modeles/boson-ai-higgs-tts-3-4b/","releve":"2026-10-19","en_bref":"25e sur 31 en synthèse vocale en français, ex æquo avec 3 autres modèles dont les marges d'erreur se recouvrent ; 12e sur 16 en prononciation du français ; 68e sur 96 en synthèse vocale en anglais (Artificial Analysis), ex æquo avec 11 autres modèles dont les marges d'erreur se recouvrent.","presentation":"Modèle de synthèse vocale de Boson AI pensé pour faire parler des agents conversationnels, servi par l'API de l'éditeur et publié avec ses poids sur Hugging Face. Il clone une voix sans entraînement et se pilote par des balises insérées dans le texte (émotion, style, débit, pauses, bruitages).","sortie":"2026-06-04","taille":"environ 4 milliards de paramètres","langues":{"annoncees":{"annoncees":"102 langues, dont le français","nombre":102,"francais":true},"prises_en_charge_selon_open_tts":102,"evaluees":["fr","en"]},"preuves":[{"appuie":["sortie"],"lu":"https://www.boson.ai/blog/higgs-tts-3 (« Higgs TTS 3: beyond reading, toward real speech for voice AI The Boson AI Team June 4, 2026 »), 17/10/2026"},{"appuie":["presentation"],"lu":"https://www.boson.ai/blog/higgs-tts-3 (« It turns model responses into expressive conversational speech across 100+ languages, with zero-shot voice cloning and inline control over emotion, style, prosody, pauses, and sound effects. »), 17/10/2026"},{"appuie":["presentation"],"lu":"https://www.boson.ai/blog/higgs-tts-3 (« For local inference, the model weights are available on Hugging Face . »), 17/10/2026"},{"appuie":["presentation"],"lu":"https://docs.boson.ai/models/higgs-tts/overview (« \"model\": \"higgs-tts-3\", \"input\": \"Hello, this is a test.\" »), 17/10/2026"},{"appuie":["langues"],"lu":"https://docs.boson.ai/models/higgs-tts/languages (« Higgs TTS 3 supports 102 languages with single-digit WER/CER. »), 17/10/2026"},{"appuie":["langues"],"lu":"https://huggingface.co/bosonai/higgs-tts-3-4b/raw/main/README.md (« The model reaches **single-digit WER/CER on 102 languages**, which split into two tiers. »), 17/10/2026"},{"appuie":["langues"],"lu":"https://huggingface.co/bosonai/higgs-tts-3-4b/raw/main/README.md (« 🇫🇷🇨🇦 French »), 17/10/2026"},{"appuie":["taille"],"lu":"https://huggingface.co/bosonai/higgs-tts-3-4b/raw/main/README.md (« | Backbone | ~4B autoregressive decoder (36 L, hidden=2560, GQA 32/8) | »), 17/10/2026"},{"appuie":["alias"],"lu":"https://huggingface.co/api/models/bosonai/higgs-audio-v3-tts-4b (« \"_id\":\"6a210f1c9a4c02e5012e4e56\",\"id\":\"bosonai/higgs-tts-3-4b\" »), 17/10/2026"},{"appuie":["alias"],"lu":"https://huggingface.co/bosonai/higgs-tts-3-4b/raw/main/README.md (« @misc{bosonai_higgs_audio_tts_v3_2026, »), 17/10/2026"}],"classements":[{"id":"synthese/francais","nom":"Synthèse vocale en français","rang":25,"sur":31,"valeur":1049.84,"unite":"Elo","bas":1028.84,"haut":1070.84,"votes":1149,"chiffres":{"prix_par_million_caracteres":15,"date_sortie":"2026-06-04"},"voisins":["fish-audio-openaudio-s1-mini","fish-audio-s2-pro"],"source":{"nom":"Artificial Analysis","mention":"Source : Artificial Analysis","url":"https://artificialanalysis.ai/text-to-speech/model-families/higgs-audio","releve":"2026-10-19"}},{"id":"synthese/prononciation","nom":"Prononciation du français","rang":12,"sur":16,"valeur":10.68,"unite":"%","chiffres":{"rtfx":50.92,"taille_milliards":4.65,"langues":102,"clonage":true},"voisins":["pocket-tts","microsoft-vibevoice-realtime-0-5b"],"source":{"nom":"Open TTS Leaderboard","mention":"Source : Open TTS Leaderboard, Hugging Face","url":"https://huggingface.co/spaces/hf-audio/open_tts_leaderboard","releve":"2026-10-19"}},{"id":"synthese/artificial-analysis","nom":"Synthèse vocale en anglais, Artificial Analysis","rang":68,"sur":96,"valeur":1041.67,"unite":"Elo","bas":1028.67,"haut":1054.67,"votes":2159,"chiffres":{"prix_par_million_caracteres":15,"date_sortie":"2026-06-04"},"voisins":["fish-audio-openaudio-s1-mini","bland-speech-v3"],"source":{"nom":"Artificial Analysis","mention":"Source : Artificial Analysis","url":"https://artificialanalysis.ai/text-to-speech/model-families/higgs-audio","releve":"2026-10-19"}}],"noms_dans_les_sources":{"artificial_analysis":"Higgs Audio V3 TTS","hf_audio":"bosonai/higgs-tts-3-4b"}}