Decentralizing Feature Extraction with Quantum Convolutional Neural Network for Automatic Speech Recognition

Abstract: We propose a novel decentralized feature extraction approach in federated learning to address privacy-preservation issues for speech recognition. It is built upon a quantum convolutional neural network (QCNN) composed of a quantum circuit encoder for feature extraction, and a recurrent neural network (RNN) based end-to-end acoustic model (AM). To enhance model parameter protection in a decentralized architecture, an input speech is first up-streamed to a quantum computing server to extract Mel-spectrogram, and the corresponding convolutional features are encoded using a quantum circuit algorithm with random parameters. The encoded features are then down-streamed to the local RNN model for the final recognition. The proposed decentralized framework takes advantage of the quantum learning progress to secure models and to avoid privacy leakage attacks. Testing on the Google Speech Commands Dataset, the proposed QCNN encoder attains a competitive accuracy of 95.12% in a decentralized model, which is better than the previous architectures using centralized RNN models with convolutional features. We also conduct an in-depth study of different quantum circuit encoder architectures to provide insights into designing QCNN-based feature extractors. Neural saliency analyses demonstrate a correlation between the proposed QCNN features, class activation maps, and input spectrograms. We provide an implementation for future studies.

• The paper introduces a decentralized QCNN approach that achieves 95.12% ASR accuracy while enhancing privacy in a federated learning framework.
• It employs a quantum circuit encoder to extract Mel-spectrogram features, which are then processed locally by an RNN for secure speech recognition.
• The results demonstrate that integrating quantum learning with classic neural networks can significantly mitigate privacy risks in acoustic data processing.

Decentralizing Feature Extraction with Quantum Convolutional Neural Networks for Automatic Speech Recognition

This paper introduces an innovative approach to enhance privacy-preserving automatic speech recognition (ASR) systems through the integration of quantum convolutional neural networks (QCNNs) within a federated learning framework. The research addresses the critical issue of acoustic data privacy and proposes a decentralized feature extraction mechanism, leveraging the strengths of Quantum Machine Learning (QML) and modern quantum technologies to ensure data protection and secure computation.

The authors present a QCNN architecture composed of a quantum circuit encoder for feature extraction and a recurrent neural network (RNN) based end-to-end acoustic model for ASR. The system operates by transmitting the input speech data to a quantum computing server to extract Mel-spectrograms and execute QCNN-based feature encoding with random parameters for enhanced security. The encoded features are then processed locally by RNN models for speech recognition. This architecture effectively utilizes quantum learning to protect model parameters and mitigate privacy leakage risks.

The empirical evaluation of the proposed system, particularly on the Google Speech Commands Dataset, showcases its efficacy, achieving a significant 95.12% accuracy in a decentralized framework—a noteworthy improvement over traditional centralized RNN-based models. This performance metric underscores the potential for QCNNs to not only enhance privacy but also maintain competitive, if not superior, accuracy in ASR applications.

Furthermore, the paper explores a meticulous exploration of various QCNN architectures to optimize feature extractor design, discussing the implications of different quantum circuit encoders on performance and providing neural saliency analysis insights into the relationship between QCNN features, class activation maps, and input Mel-spectrograms.

The practical implications of this research are profound, especially in domains where acoustic data privacy is a paramount concern, such as healthcare, finance, and personal voice assistants. Theoretically, the integration of quantum computing within traditional neural network architectures opens new avenues for research, presenting a scalable path to more robust federated learning models that capitalize on quantum advantages like parameter encryption and isolation.

In terms of future developments, this study lays the groundwork for expanding QCNN capabilities to larger, more complex ASR systems, including continuous speech recognition tasks. The authors also suggest extending the statistical privacy measures beyond the architectural decentralization offered by the QCNN models to ensure comprehensive compliance with evolving data protection standards and regulations.

This work represents a significant stride in the intersection of quantum technology and neural networks, pushing the boundaries of federated learning and setting the stage for further innovations in privacy-preserving automatic speech recognition systems.