返回
FPGA Implementation of PoolFormer Network Using Python-Driven High-Level Synthesis Framework for Edge-AIoT Speech Recognition
DOI:10.1109/TVLSI.2025.3614721.png)
摘要
En 中文
本报告介绍了一种基于新型脉冲特征提取(SFE)方法和针对现场可编程门阵列(FPGA)硬件实现优化的PoolFormer(PF)神经网络边缘AIoT语音识别系统。采用Python驱动的高层次综合(HLS)流程,以加速软件到硬件的转换进行快速验证,展示了基于FPGA的解决方案在边缘应用中的潜力。本研究为超低功耗语音识别提供了一种全面的端到端解决方案,利用HLS弥合软件开发与硬件开发之间的差距。在Xilinx PYNQ-Z2 FPGA开发板上实现时,该优化的PF模型在35类Google Commands数据集上达到了95.41%的语音识别准确率,参数量为39k。
Keyword:
Edge artificial intelligence of things (AIoT)
high-level synthesis (HLSs)
PoolFormer (PF)
PYNQ
speech recognition
期刊
I
IF:
3.1
论文数:
499
被引数:
7.3K
机构
引用论文
DWT-PoolFormer: Discrete Wavelet Transform-based Quantized Parallel PoolFormer Network Implemented in FPGA for Wearable ECG MonitoringCao, T., Ng, W.S., Goh, W.L., and Gao, Y. (2024). DWT-PoolFormer: 基于离散小波变换的量化并行PoolFormer网络在FPGA上的实现,用于可穿戴ECG监测。IEEE生物医学电路与系统会议(BioCAS)(中国西安),1-5。https://doi.org/10.1109/BioCAS61083.2024.10798386.
Edge PoolFormer: Modeling and Training of PoolFormer Network on RRAM Crossbar for Edge-AI Applications曹, T., , 余, W., 高, Y., 刘, C., 张, T., 严, S., Goh, W.L. (2025). Edge PoolFormer: RRAM交叉阵列上PoolFormer网络的建模与训练,面向Edge-AI应用. IEEE超大规模集成电路系统汇刊. 33, 384-394. https://doi.org/10.1109/TVLSI.2024.3472270.
Comparison of parametric representations for monosyllabic word recognition in continuously spoken sentences连续口语句子中单音节单词识别的参数表示形式的比较

