Integrating Quantum-Assisted Computation into Deep Speech Enhancement Networks
Long Chen, Yuezhong Tang
Abstract
Deep learning models based on Long Short-Term Memory (LSTM) architectures, such as the dual-path transformation LSTM network (DTLN), have achieved state-of-the-art performance in speech enhancement. In this work, we propose a quantum-assisted modification of DTLN, where key linear operations are replaced with quantum-inspired transformations simulated via Qiskit. By leveraging quantum state superposition and rotation-based feature interaction, the proposed approach demonstrates the feasibility of integrating quantum computation principles into classical deep speech enhancement networks. Preliminary experiments on representative noisy speech confirm that the quantum-assisted model can produce enhanced outputs consistent with the original DTLN, while maintaining compatibility with existing TensorFlow implementations and supporting end-to-end training. This work opens new avenues for exploring hybrid classical-quantum architectures in speech processing.