QC-GAN: A Parameter-Efficient Quaternion Conformer GAN for High-Fidelity Speech Enhancement

The QC-GAN framework introduces a parameter-efficient approach for speech enhancement by leveraging a Quaternion Conformer generator and MetricGAN-based training, achieving high fidelity with only 0.89 million parameters. It attained a PESQ score of 3.48 on the VoiceBank+DEMAND dataset, demonstrating performance comparable to state-of-the-art models at a fraction of the parameter count. Additionally, a smaller variant with 35K parameters achieved a PESQ score of 3.23, indicating significant efficiency improvements for practitioners focused on resource-constrained applications in speech processing.