A High-Fidelity Speech Super Resolution Network using a Complex Global Attention Module with Spectro-Temporal Loss
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866908428778078208 |
|---|---|
| author | Tamiti, Tarikul Islam Joshi, Biraj Hasan, Rida Hasan, Rashedul Athay, Taieba Mamun, Nursad Barua, Anomadarshi |
| author_facet | Tamiti, Tarikul Islam Joshi, Biraj Hasan, Rida Hasan, Rashedul Athay, Taieba Mamun, Nursad Barua, Anomadarshi |
| contents | Speech super-resolution (SSR) enhances low-resolution speech by increasing the sampling rate. While most SSR methods focus on magnitude reconstruction, recent research highlights the importance of phase reconstruction for improved perceptual quality. Therefore, we introduce CTFT-Net, a Complex Time-Frequency Transformation Network that reconstructs both magnitude and phase in complex domains for improved SSR tasks. It incorporates a complex global attention block to model inter-phoneme and inter-frequency dependencies and a complex conformer to capture long-range and local features, improving frequency reconstruction and noise robustness. CTFT-Net employs time-domain and multi-resolution frequency-domain loss functions for better generalization. Experiments show CTFT-Net outperforms state-of-the-art models (NU-Wave, WSRGlow, NVSR, AERO) on the VCTK dataset, particularly for extreme upsampling (2 kHz to 48 kHz), reconstructing high frequencies effectively without noisy artifacts. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_00229 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | A High-Fidelity Speech Super Resolution Network using a Complex Global Attention Module with Spectro-Temporal Loss Tamiti, Tarikul Islam Joshi, Biraj Hasan, Rida Hasan, Rashedul Athay, Taieba Mamun, Nursad Barua, Anomadarshi Sound Artificial Intelligence Audio and Speech Processing Speech super-resolution (SSR) enhances low-resolution speech by increasing the sampling rate. While most SSR methods focus on magnitude reconstruction, recent research highlights the importance of phase reconstruction for improved perceptual quality. Therefore, we introduce CTFT-Net, a Complex Time-Frequency Transformation Network that reconstructs both magnitude and phase in complex domains for improved SSR tasks. It incorporates a complex global attention block to model inter-phoneme and inter-frequency dependencies and a complex conformer to capture long-range and local features, improving frequency reconstruction and noise robustness. CTFT-Net employs time-domain and multi-resolution frequency-domain loss functions for better generalization. Experiments show CTFT-Net outperforms state-of-the-art models (NU-Wave, WSRGlow, NVSR, AERO) on the VCTK dataset, particularly for extreme upsampling (2 kHz to 48 kHz), reconstructing high frequencies effectively without noisy artifacts. |
| title | A High-Fidelity Speech Super Resolution Network using a Complex Global Attention Module with Spectro-Temporal Loss |
| topic | Sound Artificial Intelligence Audio and Speech Processing |
| url | https://arxiv.org/abs/2507.00229 |