TTS-1 Technical Report
Abstract
We introduce Inworld TTS-1, a set of two Transformer-based autoregressive text-to-speech (TTS) models. Our largest model, TTS-1-Max, has 8.8B parameters and is designed for utmost quality and expressiveness in demanding applications. TTS-1 is our most efficient model, with 1.6B parameters, built for real-time speech synthesis and on-device use cases. By scaling train-time compute and applying a sequential process of pre-training, fine-tuning, and RL-alignment of the speech-language model (SpeechLM) component, both models achieve state-of-the-art performance on a variety of benchmarks, demonstrating exceptional quality relying purely on in-context learning of the speaker's voice. Inworld TTS-1 and TTS-1-Max can generate high-resolution 48 kHz speech with low latency, and support 11 languages with fine-grained emotional control and non-verbal vocalizations through audio markups. We additionally open-source our training and modeling code under an MIT license.
Cite
@article{arxiv.2507.21138,
title = {TTS-1 Technical Report},
author = {Oleg Atamanenko and Anna Chalova and Joseph Coombes and Nikki Cope and Phillip Dang and Zhifeng Deng and Jimmy Du and Michael Ermolenko and Feifan Fan and Yufei Feng and Cheryl Fichter and Pavel Filimonov and Louis Fischer and Kylan Gibbs and Valeria Gusarova and Pavel Karpik and Andreas Assad Kottner and Ian Lee and Oliver Louie and Jasmine Mai and Mikhail Mamontov and Suri Mao and Nurullah Morshed and Igor Poletaev and Florin Radu and Dmytro Semernia and Evgenii Shingarev and Vikram Sivaraja and Peter Skirko and Rinat Takhautdinov and Robert Villahermosa and Jean Wang},
journal= {arXiv preprint arXiv:2507.21138},
year = {2025}
}
Comments
20 pages, 10 figures. For associated modeling and training code, see https://github.com/inworld-ai/tts