Filters








363 Hits in 6.6 sec

Tiny Transducer: A Highly-efficient Speech Recognition Model on Edge Devices [article]

Yuekai Zhang, Sining Sun, Long Ma
2021 arXiv   pre-print
This paper proposes an extremely lightweight phone-based transducer model with a tiny decoding graph on edge devices.  ...  Finally, with only 0.9M parameters after SVD, our system could give a relative 9.1% - 20.5% improvement compared with a bigger conventional hybrid system on edge devices.  ...  CONCLUSION This paper introduces the pipeline of designing a highly compact speech recognition system for extremely low-resource edge devices.  ... 
arXiv:2101.06856v2 fatcat:bptwsmzvebgezlojczeoem3uz4

Mechanical Energy Sensing and Harvesting in Micromachined Polymer-Based Piezoelectric Transducers for Fully Implanted Hearing Systems: A Review

Rhonira Latif, Mimiwaty Mohd Noor, Jumril Yunas, Azrul Azlan Hamzah
2021 Polymers  
A dual function of sensing the sound signal and simultaneously harvesting vibration energy to power up its system can be attained from a single transducer.  ...  The revision of the available transduction mechanisms, device configuration structures and piezoelectric material characteristics reveals the advantage of adopting the polymer-based piezoelectric transducers  ...  The device is placed on a simulated eardrum membrane model where its first natural frequency is expected to be at 550 Hz.  ... 
doi:10.3390/polym13142276 fatcat:zaayfonaz5g7dlsl6kyxd7n3je

Omni-sparsity DNN: Fast Sparsity Optimization for On-Device Streaming E2E ASR via Supernet [article]

Haichuan Yang, Yuan Shangguan, Dilin Wang, Meng Li, Pierce Chuang, Xiaohui Zhang, Ganesh Venkatesh, Ozlem Kalinli, Vikas Chandra
2022 arXiv   pre-print
From wearables to powerful smart devices, modern automatic speech recognition (ASR) models run on a variety of edge devices with different computational budgets.  ...  To navigate the Pareto front of model accuracy vs model size, researchers are trapped in a dilemma of optimizing model accuracy by training and fine-tuning models for each individual edge device while  ...  INTRODUCTION End-to-End (E2E) automatic speech recognition (ASR) models have gained popularity for deployment on edge devices [1, 2] .  ... 
arXiv:2110.08352v2 fatcat:velqkmn2krhdpjjutckneq6eci

Design and Evaluation of an Ultra Low-power Human-quality Speech Recognition System

Dennis Pinto, Jose-María Arnau, Antonio González
2020 ACM Transactions on Architecture and Code Optimization (TACO)  
Automatic Speech Recognition (ASR) has experienced a dramatic evolution since pioneer development of Bell Lab's single-digit recognizer more than 50 years ago.  ...  The software is based on the so-called hybrid approach with a vocabulary of 200K words and RNN-based language model re-scoring, whereas the hardware consists of a commercially available low-power processor  ...  CONCLUSIONS Automatic Speech Recognition is becoming a key technology for a large variety of computing devices.  ... 
doi:10.1145/3425604 fatcat:73qovytiebbddjvjrfce5igvny

Sensing, Computing, and Communication for Energy Harvesting IoTs: A Survey [article]

Dong Ma, Guohao Lan, Mahbub Hassan, Wen Hu, Sajal K. Das
2019 arXiv   pre-print
Next it reviews the advancements in efficient checkpointing and timekeeping for intermittently powered IoT devices.  ...  To overcome this problem, energy harvesting offers a viable alternative to autonomously power IoT devices, resulting in a number of battery-less energy harvesting IoTs (or EH-IoTs) appearing in the market  ...  More research on energy-aware debugging is required to realize highly efficient and reliable debugging and development support for EH-IoTs. E.  ... 
arXiv:1905.03949v2 fatcat:xqzuk3c6mjf5bpagdtkfdsro6e

International Journal of Computer Science July 2021

IJCSIS Editor
2021 Zenodo  
Index Terms-Deep learning, Edge accelerator, Edge computing, IOT, Tiny-YOLOv3, Traffic Sign Recognition, Edge AI, Sipeed MAIX. I.  ...  Problem statement We are implementing a deep CNN model based on Tiny-YOLOv3 on the EDGE AI Sipeed MAIX platform to solve the detection problem on the Belgium Traffic Sign dataset.  ...  Other activities that can follow home entry are KEN, REN, HEX, SNG, DNG, CAY as seen on the table 1. The current activity is shifted to 'Previous1Activity' once a new activity is predicted.  ... 
doi:10.5281/zenodo.5543977 fatcat:vxk2ynl3l5bbpcvpda2lt4qef4

Adsorption and diffusion on a stepped surface: Atomic hydrogen on Pt(211)

R. A. Olsen, Ş. C. Bădescu, S. C. Ying, E. J. Baerends
2004 Journal of Chemical Physics  
As a result, performance of speech recognition is improved. 1pSC27. Noisy speech recognition based on codebook normalization of discrete-mixture hidden Markov models.  ...  One case uses one hydrophone to measure the total sound radiated by all the droplets. The other case is based on a grid of tiny hydrophones, each measuring a droplet impact sound.  ... 
doi:10.1063/1.1755664 pmid:15268219 fatcat:qgbl7hbggvan7eyuyiow2d4yhy

A transparency model and its applications for simulation of reflector arrays and sound transmission

Claus Lynge Christensen, Jens Holger Rindel
2006 Journal of the Acoustical Society of America  
As a result, performance of speech recognition is improved. 1pSC27. Noisy speech recognition based on codebook normalization of discrete-mixture hidden Markov models.  ...  One case uses one hydrophone to measure the total sound radiated by all the droplets. The other case is based on a grid of tiny hydrophones, each measuring a droplet impact sound.  ...  Holmes ͑Boston Univ., Boston, MA 02215͒, and Scott Glenn ͑Rutgers Univeristy, NB, NJ 08901-8521͒ The feasibility of using a single transducer element on an ocean glider to create a synthetic aperture is  ... 
doi:10.1121/1.4786982 fatcat:uh2zucni3zaa5nnkivc5sh5tpq

Auralization of an orchestra using multichannel and multisource technique

Michelle C. Vigeant, Lily M. Wang, Jens Holger Rindel
2006 Journal of the Acoustical Society of America  
As a result, performance of speech recognition is improved. 1pSC27. Noisy speech recognition based on codebook normalization of discrete-mixture hidden Markov models.  ...  One case uses one hydrophone to measure the total sound radiated by all the droplets. The other case is based on a grid of tiny hydrophones, each measuring a droplet impact sound.  ...  Holmes ͑Boston Univ., Boston, MA 02215͒, and Scott Glenn ͑Rutgers Univeristy, NB, NJ 08901-8521͒ The feasibility of using a single transducer element on an ocean glider to create a synthetic aperture is  ... 
doi:10.1121/1.4787034 fatcat:brilcvuxpbgdvpdfzwovehoary

Acoustic concerns related to multi cultural societies

Anders Chr. Gade
2001 Journal of the Acoustical Society of America  
It will be explained how a no-stack thermoacoustic device may in principle have a greater efficiency than a stack-based device.  ...  In the present device the actuator is coupled instead to a flat, compliant diaphragm supported on the edges by an inflatable tube.  ...  In particular, it focused on whether listeners activate native candidate words during the recognition of non-native words.  ... 
doi:10.1121/1.4777092 fatcat:mu53dmjeund3zg5guqdnjpl2bm

Discrimination of short speech‐like formant transitions

Astrid van Wieringen, Louis C. W. Pols
1992 Journal of the Acoustical Society of America  
The speech MLD was obtained for the recognition of spondaic words masked by speech-shaped noise.  ...  A low-frequency one-sided flextensional transducer.  ...  signal in speech recognition.  ... 
doi:10.1121/1.405128 fatcat:dbtewzqvijhkta6op7ul3ribti

Machine Learning and Fuzzy Logic in Electronics: Applying Intelligence in Practice

Malinka Ivanova, Petya Petkova, Nikolay Petkov
2021 Electronics  
The applied research methodology includes a bibliographic approach in combination with a detailed examination of 66 selected papers.  ...  The findings reveal the gradually increasing interest over the last 10 years in the machine learning and fuzzy logic techniques for modeling, implementing and improving different hardware-based intelligent  ...  The proposed method for end-to-end speech recognition leads to simplifying the traditional complex process of speech recognition.  ... 
doi:10.3390/electronics10222878 fatcat:26fop4exengm5pmaf6y23ihjoe

Designing an integrated driver assistance system using image sensors

M. Akhlaq, Tarek R. Sheltami, Bo Helgeson, Elhadi M. Shakshuki
2012 Journal of Intelligent Manufacturing  
Advanced driver assistance system (ADAS) can reduce the human errors by keeping an eye on the driving environment and warning a driver to the upcoming danger.  ...  It turned out that information on speed, distance, relative position, direction, and size and type of the nearby objects would be useful and enough for implementing most of the ADAS functions.  ...  Specialized devices are also being introduced which can efficiently process visual data.  ... 
doi:10.1007/s10845-011-0618-1 fatcat:euyozj7nyvcltg2kwsbwjj7zca

Modeling and experiments with low‐frequency pressure wave propagation in liquid‐filled, flexible tubes

Cato Bjelland, Leif Bjo/rno/
1992 Journal of the Acoustical Society of America  
The speech MLD was obtained for the recognition of spondaic words masked by speech-shaped noise.  ...  A low-frequency one-sided flextensional transducer.  ...  signal in speech recognition.  ... 
doi:10.1121/1.404777 fatcat:xhmwz65h5bbqxbt52khae2rq7q

Treatment of early and late reflections in a hybrid computer model for room acoustics

Graham Naylor
1992 Journal of the Acoustical Society of America  
The speech MLD was obtained for the recognition of spondaic words masked by speech-shaped noise.  ...  A low-frequency one-sided flextensional transducer.  ...  signal in speech recognition.  ... 
doi:10.1121/1.404930 fatcat:xeehcxepjvhuzmnaeudpqxtnvu
« Previous Showing results 1 — 15 out of 363 results