---
title: "Agogic：用表现力计时音乐分词法提升文本到符号音乐生成"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.03999v1"
permalink: "https://feed.astralor.com/items/itm_256fee446ba5ddb924ccd4b6"
published: "2026-08-04T17:56:49.000Z"
score: 77
channel: "ai"
tags: ["PMT","Qwen3.5","文本到音乐生成","符号音乐","音乐分词"]
---

# Agogic：用表现力计时音乐分词法提升文本到符号音乐生成

## AI 点评

> 这篇的关键不在又刷了多少指标，而是把“分词方式到底重不重要”从口号变成可测量的事实。对做音乐AIGC的团队来说，先别急着堆参数，换个表示可能比把模型加倍更划算；对做评测的人来说，它的语料和诊断工具也值得直接拿来用。

## 速览

该论文在固定模型、数据、预算和解码条件下，只替换音乐表示方式，系统比较七种分词法对文本到符号音乐LLM的影响。结果显示，表示方式而非模型规模决定分布保真度：将主模型扩大34倍几乎不改变FMD，切换表示却能将FMD减半。作者提出的PMT性能分辨率流（10毫秒计时、逐音符力度、多轨纹理，609个符号）在0.8B规模下达到FMD 159，优于27B节拍网格模型。研究还发现标题遵循性弱但可分离，并发布训练框架、25+检查点与两个大规模语料库。

## 关键点

- 固定其他变量后仅更换音乐表示，发现表示方式比模型规模更能决定生成分布保真度；模型扩大34倍对FMD影响甚微，换表示可减半。
- 新推出的PMT性能分辨率分词（10ms计时、逐音符力度、多轨、609符号）在0.8B下FMD 159，超越27B节拍网格模型，并在26M从零训练模型上复现。
- 标题遵循能力弱但可分离：解码时轻量约束将乐器F1从.28提升到.60，调性准确率从.16提升到.35，且不损失分布质量。
- 发布完整评测框架、25+检查点、86.6k对齐语料与6.25M带标题大规模语料库，并提供一个暴露现有系统“标题印记”的诊断工具。

## 全文

Text-to-music language models begin with a choice usually made by default: how to tokenize music. Normally entangled with backbone, data, and recipe, its effect has never been measured in isolation. We fix pretrained Qwen3.5 (0.8B-27B), data, budget, and decoding, and swap only the representation across seven tokenizations, anchoring texture metrics to each representation's model-free ceiling. The ordering is clean and surprising: representation, not model size, is the binding variable for distributional fidelity. Scaling the backbone 34x barely moves Frechet Music Distance (FMD), whereas switching representation halves it. PMT, a performance-resolution stream we release (10 ms timing, per-note velocity, multi-track texture; 609 symbols), reaches FMD 159 at 0.8B against 272-286 for beat grids (1.7-1.8x lower, up to 2.8x elsewhere; non-overlapping bootstrap CIs), so a 0.8B performance-resolution model beats a 27B beat grid. It reappears on a 26M from-scratch backbone and a second performance-resolution tokenizer: a property of the class, not one lucky vocabulary. Nor is it a finer-lattice artifact: snapping PMT's onsets to the beat grids' resolution still leaves it 67-129 FMD ahead of both (n=500). The effect is distributional; whether it is audible is a separate question, left open by our probe, with a human study pre-registered. Native caption adherence is weak but separable: a lightweight decode-time constraint doubles instrument-F1 (.28 to .60) and Correct-Key (.16 to .35) at no distributional cost. We release the harness, 25+ checkpoints, two corpora (86.6k aligned across caption/MIDI/ABC/audio; 6.25M captioned, the largest for music), and an imprinting diagnostic: published text-to-MIDI systems reproduce their training distribution near-invariant to the caption (72% vs. 71% chord-time on disjoint domains). The field's next representation claim can now be measured, not asserted.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_256fee446ba5ddb924ccd4b6