SPADE让大模型语音合成更高效，保持音质与发音人特征。
SPADE: Structured Pruning and Adaptive Distillation for Efficient LLM-TTS
- 通过词错误率评估层重要性，剪枝非必要注意力层。
- 多层级知识蒸馏恢复自回归连贯性，深度减半仍保音质。
- 仅需原数据5%即可训练，适合部署于资源受限设备。
本文提出SPADE框架，用于高效的大语言模型驱动语音合成（LLM-TTS）。近期的LLM-TTS系统虽具强可控性与零样本泛化能力，但参数量大、延迟高，限制实际部署。SPADE结合两项技术：(i) 基于词错误率的层重要性指标引导结构化剪枝，移除冗余Transformer层；(ii) 多层级知识蒸馏，重建自回归生成连贯性。在零样本基准测试中，SPADE实现近似原始感知质量，将变换器深度减半，显存占用降低最多20%，实时因子提升至1.7倍，且仅需不到5%的原始训练数据。结果表明，紧凑型LLM-TTS模型可在保持自然度与说话人相似性的前提下，实现实用级实时语音生成。音频样例见https://mm.kaist.ac.kr/projects/SPADE/。
原文摘要 · Abstract (English)
The goal of this paper is to introduce SPADE, a framework for Structured Pruning and Adaptive Distillation for Efficient Large Language Model-based text-to-speech (LLM-TTS). Recent LLM-TTS systems achieve strong controllability and zero-shot generalization, but their large parameter counts and high latency limit real-world deployment. SPADE addresses this by combining (i) a pruning step guided by a word-error-rate-based layer importance index to remove non-essential Transformer layers, with (ii) multi-level knowledge distillation to restore autoregressive coherence. On zero-shot benchmarks, SPADE preserves near-parity perceptual quality while halving Transformer depth, reducing VRAM usage by up to 20%, and achieving up to 1.7x faster real-time factor with less than 5% of the original training data. These results show that compact LLM-TTS models can maintain naturalness and speaker similarity while enabling practical real-time speech generation. Audio samples are available at https://mm.kaist.ac.kr/projects/SPADE/.
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。