---
title: "Dolphin-CN-Dialect: Where Chinese Dialects Matter"
canonical_url: "https://www.modelscope.cn/papers/278861"
md_url: "https://www.modelscope.cn/papers/278861.md"
arxiv_id: 2605.08961
published: 2026-05-09
last_updated: 2026-05-09
authors:
  - "Yangyang Meng"
  - "Huihang Zhong"
  - "Guodong Lin"
  - "Guanbo Wang"
  - "Hu Du"
  - "Zhiming Shao"
  - "Yukai Huang"
  - "Ke Li"
  - "Wei-Qiang Zhang"
model_name: Dolphin-CN-Dialect
model_developer: "Dataocean AI、清华大学"
domain:
  - "语音识别"
  - "自然语言处理"
  - "方言识别"
  - "流式语音识别"
type:
  - "语音识别"
  - "自然语言处理"
  - "方言识别"
  - "流式语音识别"
  - "Computation and Language"
  - "Audio and Speech Processing"
arxiv_url: "https://arxiv.org/abs/2605.08961"
pdf_url: "https://arxiv.org/pdf/2605.08961"
---

# Dolphin-CN-Dialect: Where Chinese Dialects Matter

> We present Dolphin-CN-Dialect, a streaming-capable ASR model with a focus on Chinese and dialect-rich scenarios. Compared to the previous version, Dolphin-CN-Dialect introduces substantial improvements in data processing, tokenization, training stability,…

「Dolphin-CN-Dialect: Where Chinese Dialects Matter」是 ModelScope 魔搭社区收录的论文，arXiv 2605.08961，作者为 Yangyang Meng, Huihang Zhong, Guodong Lin et al.，发表于 2026-05-09，属于 语音识别、自然语言处理、方言识别 领域。

- **ArXiv**: 2605.08961
- **Published**: 2026-05-09
- **Authors**: Yangyang Meng, Huihang Zhong, Guodong Lin, Guanbo Wang, Hu Du, Zhiming Shao, Yukai Huang, Ke Li, Wei-Qiang Zhang
- **Model**: Dolphin-CN-Dialect
- **Developer**: Dataocean AI、清华大学
- **Domain**: 语音识别, 自然语言处理, 方言识别, 流式语音识别
- **ArXiv URL**: https://arxiv.org/abs/2605.08961
- **PDF**: https://arxiv.org/pdf/2605.08961

Source: https://www.modelscope.cn/papers/278861

---

> Dolphin-CN-Dialect：面向真实中国方言场景的轻量流式ASR引擎，用温度采样与混合分词破解数据长尾困局

## 摘要

研究背景：现有ASR模型在多方言场景下面临严重数据不平衡（标准普通话主导、方言样本稀缺）、流式部署支持不足、定制化能力弱等现实瓶颈。本文提出Dolphin-CN-Dialect——一个面向中文及丰富方言场景的流式可支持、高鲁棒性自动语音识别模型。方法上，创新性地设计了温度调节的数据采样策略以动态平衡普通话与低资源方言的训练分布；重构混合分词器：中文采用字符级建模以适配CTC-AED联合架构，英文采用BPE子词建模，并引入可扩展的方言/地域专用token（预留80个槽位）；同时构建双路径热词增强解码框架，融合编码器级上下文偏置与提示式解码偏置，兼顾稳定性与长尾词识别能力。实验结果表明：相比基线Dolphin，方言识别准确率提升38%，整体字错率（CER）降低16.3%，且在Qwen-ASR、FireredASR等SOTA开源模型中保持竞争力的同时模型体积更小；支持流式/非流式双模推理、硬件优化部署与热词实时注入。其价值在于弥合了学术模型与工业落地之间的鸿沟，为真实复杂方言环境提供了高效、轻量、可定制的ASR解决方案。

## Abstract

We present Dolphin-CN-Dialect, a streaming-capable ASR model with a focus on Chinese and dialect-rich scenarios. Compared to the previous version, Dolphin-CN-Dialect introduces substantial improvements in data processing, tokenization, training stability, and data sampling strategies. To address the challenges of highly imbalanced dialect data, we propose a temperature-based sampling strategy that effectively balances standard Mandarin and low-resource dialects, leading to significant gains in dialect recognition performance. In addition, we redesign the tokenizer to better align with linguistic characteristics, adopting character-level modeling for Chinese and subword modeling for English, while introducing extensible dialect tokens. Experimental results show that Dolphin-CN-Dialect achieves improvement in dialect recognition accuracy and CER reduction compared to Dolphin. Furthermore, Dolphin-CN-Dialect reaches competitive performance with recent SOTA open-source ASR models, while maintaining a significantly smaller model size. Dolphin-CN-Dialect supports both streaming and non-streaming inference, enabling a practical balance between latency and accuracy. It also provides flexible customization through hotword support and efficient deployment optimized for specialized hardware. These improvements make Dolphin-CN-Dialect a strong and practical solution for real-world multi-dialect ASR applications.