Scholarlink

<p>The human voice, rich in nuances such as gender, age, emotion, and language, is a powerful identifier in speaker recognition. In the rapidly growing biometrics industry, there is an increasing demand for accurate, real-time voice identification across applications like user authentication, forensic analysis, and secure banking. While traditional Deep Learning (DL) methods have advanced Speaker Identification (SI) and Speaker Verification (SV), they often struggle to isolate the unique features that distinguish a speaker’s voice from background noise in noisy environments. Additionally, the high dimensionality of extracted features, often containing redundant or irrelevant information, leads to the curse of dimensionality, making it challenging for DL models to process and identify speaker-specific patterns efficiently. To address these challenges, this study introduces a Hybrid Neural Quantum Speaker Model (HNQSM), which integrates Convolutional Neural Networks (CNN), Quantum Convolutional Neural Networks (QCNN), and a Siamese Neural Network (SNN) to enhance speaker recognition performance. This approach begins by selecting primary features from speech waveforms, allowing the model to focus on essential characteristics while reducing computational overhead. These features are compressed and refined through a CNN, then processed with quantum techniques, leveraging the efficiency and accuracy of QCNN in classification tasks. Results from the SNN layer reveal that the HNQSM outperforms state-of-the-art methods, achieving high accuracy and a low Equal Error Rate (EER). This demonstrates the model’s ability to retain the unique qualities of speech data while exploring quantum-inspired feature transformations within a hybrid architecture, establishing HNQSM as a promising advancement in voice biometrics.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

HNQSM: A Novel Hybrid Neural-Quantum Approach for Advanced Speaker Verification and Identification

  • 期刊论文

  • 发表日期：

    2026年5月22日

查看全文
  • Bhawna Jain,
  • Khushi Anand,
  • Harshita Verma,
  • Kirti Sinha

摘要

The human voice, rich in nuances such as gender, age, emotion, and language, is a powerful identifier in speaker recognition. In the rapidly growing biometrics industry, there is an increasing demand for accurate, real-time voice identification across applications like user authentication, forensic analysis, and secure banking. While traditional Deep Learning (DL) methods have advanced Speaker Identification (SI) and Speaker Verification (SV), they often struggle to isolate the unique features that distinguish a speaker’s voice from background noise in noisy environments. Additionally, the high dimensionality of extracted features, often containing redundant or irrelevant information, leads to the curse of dimensionality, making it challenging for DL models to process and identify speaker-specific patterns efficiently. To address these challenges, this study introduces a Hybrid Neural Quantum Speaker Model (HNQSM), which integrates Convolutional Neural Networks (CNN), Quantum Convolutional Neural Networks (QCNN), and a Siamese Neural Network (SNN) to enhance speaker recognition performance. This approach begins by selecting primary features from speech waveforms, allowing the model to focus on essential characteristics while reducing computational overhead. These features are compressed and refined through a CNN, then processed with quantum techniques, leveraging the efficiency and accuracy of QCNN in classification tasks. Results from the SNN layer reveal that the HNQSM outperforms state-of-the-art methods, achieving high accuracy and a low Equal Error Rate (EER). This demonstrates the model’s ability to retain the unique qualities of speech data while exploring quantum-inspired feature transformations within a hybrid architecture, establishing HNQSM as a promising advancement in voice biometrics.