HYBRID REAL- AND COMPLEX-VALUED NEURAL NETWORK CONCEPT FOR LOW-COMPLEXITY PHASE-AWARE SPEECH ENHANCEMENT
Luan Vinícius Fiorio, Alex Young, Ronald M. Aarts · 2025
Read HYBRID REAL- AND COMPLEX-VALUED NEURAL NETWORK CONCEPT FOR LOW-COMPLEXITY PHASE-AWARE SPEECH ENHANCEMENT by Luan Vinícius Fiorio, Alex Young, Ronald M. Aarts (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-21185.txt.
Description
Complex-valued neural networks handle the phase of speech naturally, but they cost more compute, especially in small models. Luan Vinicius Fiorio, Alex Young, and Ronald Aarts test a hybrid design in a 2025 paper on monaural speech enhancement. The hybrid keeps parameters matched against purely real or complex rivals.
One branch processes magnitude with a real-valued mask. A second, complex-valued branch adds a correction, and domain conversion functions couple the branches at the bottleneck. The design is applied to both a convolutional denoising autoencoder and a convolutional-recurrent network. Averaged over four noise levels, the hybrid models beat the counterparts considered on intelligibility and quality. They also use substantially fewer operations, a practical win for hearing devices and phones.
Catalog record
| CREATOR | Luan Vinícius Fiorio, Alex Young, Ronald M. Aarts |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 20 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-21185