Related references
Note: Only part of the references are listed.
Article
Engineering, Electrical & Electronic
Weiwei Shan et al.
Summary: Researchers propose a sub-microwatt keyword spotting (KWS) chip with an acoustic activity detection (AAD) to achieve ultra-low power and high detection accuracy. The chip utilizes an optimized feature extractor circuit, a zero-cost AAD, a tunable detection window, and a true form computation method. Implemented in a 28-nm CMOS process, the chip demonstrates promising performance with a 97.8% accuracy for two keywords in the Google Speech Command Dataset (GSCD).
IEEE JOURNAL OF SOLID-STATE CIRCUITS
(2023)
Article
Chemistry, Analytical
Seongwoo Bae et al.
Summary: This paper proposes a DS-BTNN hardware accelerator capable of performing both wake-up-word recognition and command classification on a single device, achieving significant area efficiency. The designed KWS system on an FPGA board achieved high accuracy in wake-up-word recognition and command classification.
Proceedings Paper
Computer Science, Hardware & Architecture
Jinsung Yoon et al.
Summary: This paper presents a real-time keyword spotting system implemented in an FPGA. The system utilizes a binary convolutional neural network (BCNN) to perform the entire KWS task without involving complicated processing. The proposed system efficiently carries out BCNN inference by skipping redundant operations. It has been implemented with only 8475 lookup tables in an FPGA and achieves a spotting accuracy of 91.64% by processing one-second frame in 19.8 ms.
2023 IEEE SYMPOSIUM IN LOW-POWER AND HIGH-SPEED CHIPS, COOL CHIPS
(2023)
Article
Engineering, Electrical & Electronic
Bo Liu et al.
Summary: This study aims to improve the accuracy of low-power keywords recognition and proposes a domain-specific speech recognition microprocessor based on 1D-CRNN. An energy-efficient accelerator is implemented to reduce power consumption while maintaining high recognition accuracy.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS I-REGULAR PAPERS
(2022)
Article
Computer Science, Artificial Intelligence
Zewen Li et al.
Summary: This review provides insights into the development history of CNN, a overview of various convolutions, introduction to classic and advanced CNN models, conclusions drawn from experimental analysis, rules of thumb for function and hyperparameter selection, and applications of 1-D, 2-D, and multidimensional convolutions. Moreover, it also discusses open issues and promising directions for CNN as guidelines for future work.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Computer Science, Information Systems
Su-Jung Lee et al.
Summary: A binary convolutional neural network (BCNN) is designed for efficient visual imagery analysis in low-cost resource-limited devices. This study introduces an inference processor named TORRES for BCNNs, which skips operations efficiently based on the spatial locality in feature maps to achieve high efficiency. The prototype system has been implemented on a 28 nm field-programmable gate array, demonstrating resource efficiency.
Article
Computer Science, Information Systems
Keyan He et al.
Summary: This paper proposes a small-footprint model based on TENet and simplified MFCC algorithm, achieving high accuracy through batch-norm folding and int8 quantization. Furthermore, an optimized dataflow and configurable hardware architecture are proposed using hardware/model co-design approach.
Proceedings Paper
Computer Science, Hardware & Architecture
Yingfeng Wang et al.
Summary: This paper proposes a noise-robust, lightweight, and accurate keyword spotting system. By training and optimizing a neural network model from scratch, high accuracy keyword spotting is achieved in various noisy environments.
2022 19TH INTERNATIONAL SOC DESIGN CONFERENCE (ISOCC)
(2022)
Article
Engineering, Electrical & Electronic
Weiwei Shan et al.
Summary: The proposed sub-mu W always-ON keyword spotting chip for audio wake-up systems utilizes various techniques to achieve ultra-low power consumption, enabling high accuracy in voice wake-up tasks.
IEEE JOURNAL OF SOLID-STATE CIRCUITS
(2021)
Article
Engineering, Electrical & Electronic
Yexin Yan et al.
Summary: The study implemented two benchmark tasks on SpiNNaker 2 and Loihi neuromorphic chips, and found that the application of MAC array on SpiNNaker 2 showed better efficiency in handling high-dimensional vector-matrix multiplication.
NEUROMORPHIC COMPUTING AND ENGINEERING
(2021)
Article
Computer Science, Hardware & Architecture
Jaehyeong Sim et al.
IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS
(2020)
Proceedings Paper
Acoustics
Enea Ceolini et al.
2019 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP)
(2019)
