BEGIN:VCALENDAR
VERSION:2.0
PRODID:IEEE vTools.Events//EN
CALSCALE:GREGORIAN
BEGIN:VTIMEZONE
TZID:Asia/Shanghai
BEGIN:STANDARD
DTSTART:19910915T010000
TZOFFSETFROM:+0900
TZOFFSETTO:+0800
TZNAME:CST
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20241127T040750Z
UID:4DED9E04-313E-42C9-83A2-67AC713485F7
DTSTART;TZID=Asia/Shanghai:20241126T100000
DTEND;TZID=Asia/Shanghai:20241126T115000
DESCRIPTION:As a widely used speech-triggered interface\, deep-learning-bas
 ed keyword spotting (KWS) chips are required for wake-up tasks. The always
 -on characteristic requires KWS to achieve ultra-low power while keeping g
 ood accuracy. This talk will show the design challenges and three case stu
 dies toward ultra-low power KWS chip design. First\, a digital Keyword-Spo
 tting Chip Using Serial-FFT-based MFCC and Binarized Depthwise Separable C
 NN\, which for the first time pushed its power below 1μW. Second\, a sub-
 μW KWS chip with an acoustic activity detection (AAD) to achieve both low
  power and decent detection accuracy. Third\, a full-system Near-Microphon
 e Keyword-Spotting Chip\, which integrated the analog front-end into the c
 hip\, along with optimized MFCC and Temporal Depthwise Separable CNN to ac
 hieve a full-system solution with only 608nW.\n\nCo-sponsored by: Huazhong
  University of Science and Technology\n\nSpeaker(s): Shan Weiwei\, \n\nRoo
 m: Meeting room 117\, Bldg: New building of SOEI\, 1037 Luoyu Road\, Wuhan
 \, Hubei\, P.R.China\, 430074\, Wuhan\, Hubei\, China\, 430074
LOCATION:Room: Meeting room 117\, Bldg: New building of SOEI\, 1037 Luoyu R
 oad\, Wuhan\, Hubei\, P.R.China\, 430074\, Wuhan\, Hubei\, China\, 430074
ORGANIZER:chao_wang_me@hust.edu.cn
SEQUENCE:17
SUMMARY:IEEE CASS-EDS-SSCS & HUST Technical Talk No. 46 “ Ultra-low power
  AI accelerator design for always-on keyword spotting application”
URL;VALUE=URI:https://events.vtools.ieee.org/m/448561
X-ALT-DESC:Description: <br /><p class="MsoNormal">As a widely used speech-
 triggered interface\, deep-learning-based keyword spotting (KWS) chips are
  required for wake-up tasks. The always-on characteristic requires KWS to 
 achieve ultra-low power while keeping good accuracy. <span style="font-fam
 ily: Times New Roman\;">T</span>his talk will show the design challenges a
 nd three case studies toward ultra<span style="font-family: Times New Roma
 n\;">-</span>low power KWS chip design. First\, a digital Keyword-Spotting
  Chip Using Serial-FFT-based MFCC and Binarized Depthwise Separable CNN\, 
 which for the first time pushed its power below 1&mu\;W. &nbsp\;Second\, a
  sub-&mu\;W KWS chip with an acoustic activity detection (AAD) to achieve 
 both low power and decent detection accuracy. Third\, a full-system Near-M
 icrophone Keyword-Spotting Chip\, which integrated the analog front-end in
 to the chip\, along with optimized MFCC and Temporal Depthwise Separable C
 NN to achieve a full-system solution with only 608nW.</p>
END:VEVENT
END:VCALENDAR