用1.58比特量化+权重索引，让语音合成模型缩小83%。
BitTTS: Highly Compact Text-to-Speech Using 1.58-bit Quantization and Weight Indexing
- 训练时进行1.58比特量化，多数参数压缩为-1/0/1三值。
- 通过8位索引存储1.58比特权重，实现高效压缩。
- 模型体积减83%，音质还优于同规模无量化模型。
本文提出一种高度紧凑的轻量级文本到语音（TTS）模型，适用于设备端应用。为减小模型尺寸，引入两项技术：首先，采用量化感知训练（QAT），在训练中将模型参数量化至最低1.58比特，此时大部分32比特参数被量化为{-1, 0, 1}三值；其次，提出权重索引方法，将一组1.58比特权重以单个int8索引形式存储，实现对硬件按8比特单位处理的兼容性。实验表明，该方法使模型尺寸减少83%，同时在合成质量上优于同等规模未量化基线模型。
原文摘要 · Abstract (English)
This paper proposes a highly compact, lightweight text-to-speech (TTS) model for on-device applications. To reduce the model size, the proposed model introduces two techniques. First, we introduce quantization-aware training (QAT), which quantizes model parameters during training to as low as 1.58-bit. In this case, most of 32-bit model parameters are quantized to ternary values {-1, 0, 1}. Second, we propose a method named weight indexing. In this method, we save a group of 1.58-bit weights as a single int8 index. This allows for efficient storage of model parameters, even on hardware that treats values in units of 8-bit. Experimental results demonstrate that the proposed method achieved 83 % reduction in model size, while outperforming the baseline of similar model size without quantization in synthesis quality.
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。