Scholarlink

<p>Voice disorders pose significant challenges to an individual’s communication abilities and quality of life. Traditional methods for detecting these disorders rely on subjective assessments and manual analysis. Recent works have concentrated on using machine learning algorithms to classify voice disorders. Since these algorithms rely on extensive training data, the works have employed data augmentation and transfer learning techniques to address this requirement. However, while these approaches have not quantized their machine learning algorithms for deployment on low-resource hardware like Field Programmable Gate Array (FPGA), this paper proposes a highly efficient, lightweight Convolutional Neural Network (CNN) architecture customized for the classification of voice disorders. This is designed to be easily trainable with fewer parameters (pruning) and can be quantized for enhanced precision with QKeras. Furthermore, the quantized CNN model is converted into a high-level synthesis (HLS) project using hls4ml and synthesized with Vivado HLS to facilitate implementation on the Xilinx ZCU102 FPGA without compromising performance and achieving a classification accuracy of 99.54%.</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

Optimized FPGA Architecture for CNN-Driven Voice Disorder Detection

  • 期刊论文

  • 发表日期：

    2025年2月11日

查看全文
  • Jyoti Mishra,
  • R. K. Sharma

摘要

Voice disorders pose significant challenges to an individual’s communication abilities and quality of life. Traditional methods for detecting these disorders rely on subjective assessments and manual analysis. Recent works have concentrated on using machine learning algorithms to classify voice disorders. Since these algorithms rely on extensive training data, the works have employed data augmentation and transfer learning techniques to address this requirement. However, while these approaches have not quantized their machine learning algorithms for deployment on low-resource hardware like Field Programmable Gate Array (FPGA), this paper proposes a highly efficient, lightweight Convolutional Neural Network (CNN) architecture customized for the classification of voice disorders. This is designed to be easily trainable with fewer parameters (pruning) and can be quantized for enhanced precision with QKeras. Furthermore, the quantized CNN model is converted into a high-level synthesis (HLS) project using hls4ml and synthesized with Vivado HLS to facilitate implementation on the Xilinx ZCU102 FPGA without compromising performance and achieving a classification accuracy of 99.54%.