ZipEnhancer: Dual-Path Down-Up Sampling-based Zipformer for Monaural Speech Enhancement

2025-01-09

AI Summary

To address the high computational cost and poor scalability of dual-path time-domain–time-frequency-domain models in single-channel speech enhancement, this paper proposes Zipformer-Enhancer—a lightweight dual-path architecture leveraging downsampled and upsampled representations. Key contributions include: (i) the first symmetric Dual-Path DownSampleStacks, which drastically reduces hidden feature dimensionality; (ii) a parameter-efficient ZipformerBlock designed for enhanced representational capacity with minimal overhead; and (iii) ScaleAdam, a sparse-gradient-adapted optimizer, coupled with the Eden learning rate scheduler. Evaluated on DNS 2020 and VoiceBank+DEMAND, Zipformer-Enhancer achieves state-of-the-art performance (PESQ = 3.69 / 3.63) with only 2.04M parameters and 62.41G FLOPS—marking a significant improvement in the trade-off between computational efficiency and modeling capability.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Natural Language Processing: Learning & Optimization for NLP
• Search and Optimization: Learning to Search

Application Category

• Security and Privacy: Privacy-enhancing technologies
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines

Abstract

In contrast to other sequence tasks modeling hidden layer features with three axes, Dual-Path time and time-frequency domain speech enhancement models are effective and have low parameters but are computationally demanding due to their hidden layer features with four axes. We propose ZipEnhancer, which is Dual-Path Down-Up Sampling-based Zipformer for Monaural Speech Enhancement, incorporating time and frequency domain Down-Up sampling to reduce computational costs. We introduce the ZipformerBlock as the core block and propose the design of the Dual-Path DownSampleStacks that symmetrically scale down and scale up. Also, we introduce the ScaleAdam optimizer and Eden learning rate scheduler to improve the performance further. Our model achieves new state-of-the-art results on the DNS 2020 Challenge and Voicebank+DEMAND datasets, with a perceptual evaluation of speech quality (PESQ) of 3.69 and 3.63, using 2.04M parameters and 62.41G FLOPS, outperforming other methods with similar complexity levels.

Problem

• Speech Enhancement
• High Computational Cost
• Monoaural Audio Quality

Innovation

• ZipEnhancer
• Zipformer
• Dual-path Sampling