PulseAugur
EN
LIVE 08:40:34

New QC-GAN offers parameter-efficient speech enhancement

By PulseAugur Editorial · [3 sources] ·

Researchers have developed QC-GAN, a new parameter-efficient framework for speech enhancement that combines a Quaternion Conformer generator with MetricGAN-based training. This approach utilizes the Hamilton product to encode magnitude and phase, significantly reducing parameters while maintaining interdependencies. A metric-learning discriminator optimizes perceptual quality, achieving a PESQ score of 3.48 with only 0.89M parameters on the VoiceBank+DEMAND dataset, and a smaller variant with 35K parameters also showed strong performance. The model demonstrated generalization capabilities on the DNS-Challenge 3 dataset. AI

IMPACT This research introduces a more parameter-efficient approach to speech enhancement, potentially enabling higher-quality audio processing on devices with limited computational resources.

RANK_REASON The cluster contains an academic paper detailing a new model and methodology for speech enhancement.

Read on arXiv cs.AI →

AI-generated summary · Google Gemini · from 3 sources. How we write summaries →

New QC-GAN offers parameter-efficient speech enhancement

COVERAGE [3]

  1. arXiv cs.AI TIER_1 English(EN) · Shogo Yamauchi, Hideaki Tamori, Makoto Sakai, Yosuke Yamano, Tohru Nitta ·

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

    arXiv:2606.18611v1 Announce Type: cross Abstract: We propose a parameter-efficient speech enhancement framework, Quaternion Conformer GAN (QC-GAN), which combines a Quaternion Conformer generator with MetricGAN-based training. The Hamilton product encodes the magnitude and phase …

  2. arXiv stat.ML TIER_1 English(EN) · Tohru Nitta ·

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

    We propose a parameter-efficient speech enhancement framework, Quaternion Conformer GAN (QC-GAN), which combines a Quaternion Conformer generator with MetricGAN-based training. The Hamilton product encodes the magnitude and phase via structured weight sharing, reducing the number…

  3. arXiv stat.ML TIER_1 English(EN) · Tohru Nitta ·

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

    We propose a parameter-efficient speech enhancement framework, Quaternion Conformer GAN (QC-GAN), which combines a Quaternion Conformer generator with MetricGAN-based training. The Hamilton product encodes the magnitude and phase via structured weight sharing, reducing the number…

RELATED ENTITIES

RELATED TOPICS