---
title: "Qwen-Music Technical Report"
canonical_url: "https://www.modelscope.cn/papers/310815"
md_url: "https://www.modelscope.cn/papers/310815.md"
arxiv_id: 2607.11699
published: 2026-07-13
last_updated: 2026-07-13
authors:
  - "Jin Xu"
  - "Kangdi Wang"
  - "Ruibin Yuan"
  - "Shun Lei"
  - "Xiong Wang"
  - "Xize Cheng"
  - "Xueyao Zhang"
  - "Yang Zhang"
  - "Yiheng Chen"
  - "Yongqi Wang"
  - "Yue Wang"
  - "Zhifang Guo"
  - "Zihan Liu"
  - "Zijian Lin"
  - "Dake Guo"
  - "Hangrui Hu"
  - "Lei Xie"
  - "Linhan Ma"
  - "Wei Xue"
  - "Wenxiang Guo"
  - "Xinfa Zhu"
  - "Xipin Wei"
  - "Yangze Li"
  - "Yuanjun Lv"
  - "Yuxuan Wang"
  - "Yunfei Chu"
  - "Zhiyong Wu"
model_name: Qwen-Music
model_developer: "Qwen Team"
domain:
  - "音频生成"
  - "音乐生成"
  - "语音合成"
  - "深度学习"
type:
  - "音频生成"
  - "音乐生成"
  - "语音合成"
  - "深度学习"
  - Sound
arxiv_url: "https://arxiv.org/abs/2607.11699"
pdf_url: "https://arxiv.org/pdf/2607.11699"
---

# Qwen-Music Technical Report

> We introduce Qwen-Music, a music generation model that produces high-fidelity songs with complete vocals. It supports text-to-music generation from descriptions, lyrics, and musical attributes, and cover song generation with different styles and vocal…

「Qwen-Music Technical Report」是 ModelScope 魔搭社区收录的论文，arXiv 2607.11699，作者为 Jin Xu, Kangdi Wang, Ruibin Yuan et al.，发表于 2026-07-13，属于 音频生成、音乐生成、语音合成 领域。

- **ArXiv**: 2607.11699
- **Published**: 2026-07-13
- **Authors**: Jin Xu, Kangdi Wang, Ruibin Yuan, Shun Lei, Xiong Wang, Xize Cheng, Xueyao Zhang, Yang Zhang, Yiheng Chen, Yongqi Wang, Yue Wang, Zhifang Guo, Zihan Liu, Zijian Lin, Dake Guo, Hangrui Hu, Lei Xie, Linhan Ma, Wei Xue, Wenxiang Guo, Xinfa Zhu, Xipin Wei, Yangze Li, Yuanjun Lv, Yuxuan Wang, Yunfei Chu, Zhiyong Wu
- **Model**: Qwen-Music
- **Developer**: Qwen Team
- **Domain**: 音频生成, 音乐生成, 语音合成, 深度学习
- **ArXiv URL**: https://arxiv.org/abs/2607.11699
- **PDF**: https://arxiv.org/pdf/2607.11699

Source: https://www.modelscope.cn/papers/310815

---

> Qwen-Music 技术报告

## 摘要

Qwen-Music 是一个强大的音乐生成模型，能够生成具有高音乐性和高保真度的完整人声演唱歌曲。该模型支持两项核心任务：文本到音乐生成（根据文本描述、歌词和音乐属性创作全新歌曲）以及翻唱歌曲生成（以不同风格和声音特征重新演绎现有歌曲）。系统架构由三个核心组件构成：Qwen-Music-Tokenizer 将音频压缩为 25 Hz 单码本的音乐语义标记流；Qwen-Music-LLM 基于 Melody-CoT（旋律思维链）机制进行自回归音乐语义建模；Qwen-Music-Render 通过语义条件扩散 Transformer (DiT)、Spec-VAE 解码器和 Band-Mode Refiner 生成 48 kHz 高保真立体声波形。训练采用质量分级预训练课程与多阶段后训练对齐（SFT、DPO、GSPO）流程。

## Abstract

We introduce Qwen-Music, a music generation model that produces high-fidelity songs with complete vocals. It supports text-to-music generation from descriptions, lyrics, and musical attributes, and cover song generation with different styles and vocal characteristics. Qwen-Music comprises three components: Qwen-Music-Tokenizer, Qwen-Music-LLM, and Qwen-Music-Render. The tokenizer compresses audio into a 25 Hz single-codebook stream of Music Semantic Tokens that preserve semantic and melodic information. The LLM performs autoregressive modeling with a melody-token-based chain-of-thought (Melody-CoT) mechanism that plans melodies before full-song generation, improving musicality, structural coherence, and reference-melody preservation. The renderer enriches discrete semantic tokens with acoustic details to produce high-fidelity stereo waveforms. We train the LLM using a quality-aware pre-training curriculum followed by progressive post-training with supervised initialization, offline DPO, and online GSPO to improve musicality and instruction following. On 600 evaluation inputs, Qwen-Music achieves state-of-the-art results in 13 of 16 objective musicality and audio-quality metrics. Professional evaluators also prefer Qwen-Music over leading proprietary systems. For cover generation, Qwen-Music preserves reference melodies more accurately than Suno V5.5, Suno V5, and MiniMax Cover on the AI-generated reference set, and outperforms MiniMax Cover on most metrics on the real-world popular-song reference set.