用低秩适配器提升稀疏语言的语音合成质量，效果显著且参数极少。
Closing the Quality Gap in Low-Resource Text-to-Speech: LoRA Fine-Tuning of VoxCPM2 for Khmer and Korean
- 仅用一个共享适配器同时微调多语言模型，参数量仅占原模型的0.19%~3.03%
- 高棉语语音质量评分（MOS）从3.85提升至4.23，显著改善（p<0.001）
- 适配器对基础表现已优的语言无效，凸显其在弱项上的针对性改进
大模型在资源丰富语言上接近真人发音，但在数据稀缺语言上表现较差。本文以24亿参数、无分词器的VoxCPM2模型研究高棉语和韩语的语音合成质量差距。构建约26小时的多语言共享语料库，采用单个零初始化低秩适配器（LoRA），同时应用于语言模型与扩散解码器，实现双语联合微调。听觉测试显示，高棉语平均意见分（MOS）从3.85升至4.23（最佳适配器秩为64），提升显著（配对威尔科克森检验，p<0.001），而训练参数比例仅为0.19%~3.03%。自动损失最低出现在秩128，但人类评分峰值在秩64，二者不一致。该适配器对韩语无增益，高秩时甚至劣化，表明其主要改善基线表现较弱的语言。
原文摘要 · Abstract (English)
Large pretrained text-to-speech (TTS) models sound almost human for well-resourced languages, but much worse for languages that are rare in their training data. We study this quality gap for Khmer and Korean using VoxCPM2, a 2.4B-parameter, tokenizer-free TTS model that joins a MiniCPM-4 language-model backbone with a flow-matching diffusion decoder. We build one shared, language-tagged corpus of about 26 hours and adapt VoxCPM2 with a single Low-Rank Adaptation (LoRA) adapter, trained on both languages at once and added to both the language model and the decoder. The adapter is zero-initialized, so training starts exactly at the original (zero-shot) model. In native-speaker listening tests, the Khmer Mean Opinion Score (MOS) rises from 3.85 to 4.23 with the best adapter (rank 64), a highly significant gain (paired Wilcoxon test, p<0.001), while training only 0.19 to 3.03 percent of the parameters. The automatic loss and the human ratings, however, disagree on the best rank: validation loss is lowest at rank 128, yet MOS peaks at rank 64. The same adapter brings no gain for Korean, a language the base model already handles well, and at a high rank it even degrades quality. Adaptation therefore helps mainly where the base model is genuinely weak.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。