Voxtral TTS
Artificial Intelligence
2026-04-07 v2
Abstract
We introduce Voxtral TTS, an expressive multilingual text-to-speech model that generates natural speech from as little as 3 seconds of reference audio. Voxtral TTS adopts a hybrid architecture that combines auto-regressive generation of semantic speech tokens with flow-matching for acoustic tokens. These tokens are encoded and decoded with Voxtral Codec, a speech tokenizer trained from scratch with a hybrid VQ-FSQ quantization scheme. In human evaluations conducted by native speakers, Voxtral TTS is preferred for multilingual voice cloning due to its naturalness and expressivity, achieving a 68.4\% win rate over ElevenLabs Flash v2.5. We release the model weights under a CC BY-NC license.
Cite
@article{arxiv.2603.25551,
title = {Voxtral TTS},
author = {Mistral-AI and : and Alexander H. Liu and Alexis Tacnet and Andy Ehrenberg and Andy Lo and Chen-Yo Sun and Guillaume Lample and Henry Lagarde and Jean-Malo Delignon and Jaeyoung Kim and John Harvill and Khyathi Raghavi Chandu and Lorenzo Signoretti and Margaret Jennings and Patrick von Platen and Pavankumar Reddy Muddireddy and Rohin Arora and Sanchit Gandhi and Samuel Humeau and Soham Ghosh and Srijan Mishra and Van Phung and Abdelaziz Bounhar and Abhinav Rastogi and Adrien Sadé and Alan Jeffares and Albert Jiang and Alexandre Cahill and Alexandre Gavaudan and Alexandre Sablayrolles and Amélie Héliou and Amos You and Andrew Bai and Andrew Zhao and Angele Lenglemetz and Anmol Agarwal and Anton Eliseev and Antonia Calvi and Arjun Majumdar and Arthur Fournier and Artjom Joosen and Avi Sooriyarachchi and Aysenur Karaduman Utkur and Baptiste Bout and Baptiste Rozière and Baudouin De Monicault and Benjamin Tibi and Bowen Yang and Charlotte Cronjäger and Clémence Lanfranchi and Connor Chen and Corentin Barreau and Corentin Sautier and Cyprien Courtot and Darius Dabert and Diego de las Casas and Elizaveta Demyanenko and Elliot Chane-Sane and Emmanuel Gottlob and Enguerrand Paquin and Etienne Goffinet and Fabien Niel and Faruk Ahmed and Federico Baldassarre and Gabrielle Berrada and Gaëtan Ecrepont and Gauthier Guinet and Genevieve Hayes and Georgii Novikov and Giada Pistilli and Guillaume Kunsch and Guillaume Martin and Guillaume Raille and Gunjan Dhanuka and Gunshi Gupta and Han Zhou and Harshil Shah and Hope McGovern and Hugo Thimonier and Indraneel Mukherjee and Irene Zhang and Jacques Sun and Jan Ludziejewski and Jason Rute and Jérémie Dentan and Joachim Studnia and Jonas Amar and Joséphine Delas and Josselin Somerville Roberts and Julien Tauran and Karmesh Yadav and Kartik Khandelwal and Kilian Tep and Kush Jain and Laurence Aitchison and Laurent Fainsin and Léonard Blier and Lingxiao Zhao and Louis Martin and Lucile Saulnier and Luyu Gao and Maarten Buyl and Manan Sharma and Marie Pellat and Mark Prins and Martin Alexandre and Mathieu Poirée and Mathieu Schmitt and Mathilde Guillaumin and Matthieu Dinot and Matthieu Futeral and Maxime Darrin and Maximilian Augustin and Mert Unsal and Mia Chiquier and Mikhail Biriuchinskii and Minh-Quang Pham and Mircea Lica and Morgane Rivière and Nathan Grinsztajn and Neha Gupta and Olivier Bousquet and Olivier Duchenne and Patricia Wang and Paul Jacob and Paul Wambergue and Paula Kurylowicz and Philippe Pinel and Philomène Chagniot and Pierre Stock and Piotr Miłoś and Prateek Gupta and Pravesh Agrawal and Quentin Torroba and Ram Ramrakhya and Randall Isenhour and Rishi Shah and Romain Sauvestre and Roman Soletskyi and Rosalie Millner and Rupert Menneer and Sagar Vaze and Samuel Barry and Samuel Belkadi and Sandeep Subramanian and Sean Cha and Shashwat Verma and Siddhant Waghjale and Siddharth Gandhi and Simon Lepage and Sumukh Aithal and Szymon Antoniak and Tarun Kumar Vangani and Teven Le Scao and Théo Cachet and Theo Simon Sorg and Thibaut Lavril and Thomas Chabal and Thomas Foubert and Thomas Robert and Thomas Wang and Tim Lawson and Tom Bewley and Tom Edwards and Tyler Wang and Umar Jamil and Umberto Tomasini and Valeriia Nemychnikova and Vedant Nanda and Victor Jouault and Vincent Maladière and Vincent Pfister and Virgile Richard and Vladislav Bataev and Wassim Bouaziz and Wen-Ding Li and William Havard and William Marshall and Xinghui Li and Xingran Guo and Xinyu Yang and Yannic Neuhaus and Yassine El Ouahidi and Yassir Bendou and Yihan Wang and Yimu Pan and Zaccharie Ramzi and Zhenlin Xu},
journal= {arXiv preprint arXiv:2603.25551},
year = {2026}
}