BitTTS: Highly Compact Text-to-Speech Using 1.58-bit Quantization and Weight Indexing

TL;DR

This paper introduces a highly compact TTS model using ultra-low 1.58-bit quantization and weight indexing, achieving significant size reduction while maintaining high synthesis quality for on-device applications.

Contribution

The paper presents a novel combination of quantization-aware training and weight indexing to drastically reduce TTS model size without sacrificing quality.

Findings

Achieved 83% reduction in model size.

Outperformed baseline models in synthesis quality.

Demonstrated effectiveness on on-device TTS applications.

Abstract

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.