Vision & Multimodal Jan 2026 · #60 most cited · 88 citations

Qwen3-TTS Technical Report

Hangrui Hu et al.

arXiv:2601.15621

In short

Qwen3-TTS is an open, multilingual text-to-speech family that can clone a voice from three seconds of audio or design one from a description. Trained on over five million hours of speech, it streams its first audio in under 100 milliseconds using two custom speech tokenizers.

Why it matters

State-of-the-art, controllable open speech synthesis, released under Apache 2.0.

Read first

The 3 Field Guide ideas this paper leans on.

Vision & Multimodal · Reference Text-to-Speech Synthesizing natural-sounding speech from text, using neural vocoders and attention-based models. Language & LLMs · Landmark Token The basic unit of text that a language model processes, typically representing a word, subword, or character. Tokens are the fundamental building blocks for LLM input and output. Language & LLMs · Landmark Language Modeling Learning probability distributions over sequences of words to predict what comes next.

Starting from scratch? The full route 9 ideas · basics first

Dataset ✓ understood
A collection of data examples used for training, validating, or testing machine learning models.
Feature ✓ understood
A single measurable property of an example, such as a house's floor area or how many links an email contains, used as an input to a model.
Feature Engineering ✓ understood
The process of selecting, transforming, and creating input features to improve model performance.
Audio Processing ✓ understood
Techniques for analyzing, transforming, and understanding audio signals for tasks like speech recognition and music generation.
Text-to-Speech · read first ✓ understood
Synthesizing natural-sounding speech from text, using neural vocoders and attention-based models.
Token · read first ✓ understood
The basic unit of text that a language model processes, typically representing a word, subword, or character. Tokens are the fundamental building blocks for LLM input and output.
Natural Language Processing ✓ understood
The field of AI that lets computers read, interpret, translate and generate human language, from spam filters and search to chatbots.
Tokenization ✓ understood
Splitting text into tokens, usually subword pieces, and mapping each to an integer ID so a language model can process it.
Language Modeling · read first ✓ understood
Learning probability distributions over sequences of words to predict what comes next.

In the frontier

Rank: #60 of 100
Citations: 88; as of Aug 9, 2026
Published: Jan 2026

Topics: Audio, speech, and omni-modal , Model architecture

Selection: 1kpapers.com by Together AI, most-cited as of Aug 9, 2026

Nearby papers

Canon · 2020 Language Models are Few-Shot Learners It showed that scale alone unlocks in-context learning, the capability today’s prompt-driven AI is built on. Frontier · #9 Qwen3-Omni Technical Report It shows a single open model can match specialist models across every modality at once. Frontier · #12 Why Language Models Hallucinate It reframes hallucination as an incentive problem baked into how we measure models, not a mystery. Frontier · #25 Seed Diffusion: A Large-Scale Diffusion Language Model with High-Speed Inference It is evidence that diffusion language models can be dramatically faster without giving up quality.