%PDF-1.3 1 0 obj << /Kids [ 3 0 R 4 0 R 5 0 R 6 0 R 7 0 R 8 0 R 9 0 R 10 0 R 11 0 R 12 0 R ] /Type /Pages /Count 10 >> endobj 2 0 obj << /Title (Lip Reading Sentences in the Wild) /Producer (PyPDF2) /Author (Joon Son Chung\054 Andrew Senior\054 Oriol Vinyals\054 Andrew Zisserman) /Subject (2017 IEEE Conference on Computer Vision and Pattern Recognition) >> endobj 3 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 14 0 R /Resources << /XObject << /x8 15 0 R /x6 18 0 R /x12 21 0 R /x10 24 0 R >> /ExtGState << /s9 27 0 R /s11 30 0 R /R27 33 0 R /a0 << /CA 1 /ca 1 >> /s5 34 0 R /s7 37 0 R >> /Font << /F2 40 0 R /R28 41 0 R /F1 44 0 R /R38 45 0 R /R36 49 0 R /R34 53 0 R /R32 56 0 R /R30 59 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 63 0 R 64 0 R 65 0 R 66 0 R 67 0 R 68 0 R 69 0 R 70 0 R 71 0 R 72 0 R 73 0 R 74 0 R 75 0 R 76 0 R 77 0 R ] >> endobj 4 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 78 0 R /Resources << /ExtGState << /R27 33 0 R >> /Font << /R79 79 0 R /R73 83 0 R /R71 86 0 R /R77 90 0 R /R75 93 0 R /F2 96 0 R /R28 41 0 R /F1 97 0 R /R85 98 0 R /R38 45 0 R /R83 101 0 R /R81 105 0 R /R30 59 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 109 0 R 110 0 R 111 0 R 112 0 R 113 0 R 114 0 R 115 0 R 116 0 R 117 0 R 118 0 R 119 0 R 120 0 R 121 0 R 122 0 R 123 0 R 124 0 R 125 0 R 126 0 R 127 0 R 128 0 R 129 0 R 130 0 R 131 0 R 132 0 R 133 0 R 134 0 R ] >> endobj 5 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 135 0 R /Resources << /XObject << /R98 136 0 R /R99 137 0 R /R120 138 0 R /R101 141 0 R /R100 142 0 R /R103 143 0 R /R102 144 0 R >> /ExtGState << /R27 33 0 R >> /Font << /R79 79 0 R /R73 83 0 R /R71 86 0 R /R77 90 0 R /R75 93 0 R /R127 145 0 R /R125 148 0 R /F1 152 0 R /R123 153 0 R /R81 105 0 R /R85 98 0 R /R108 156 0 R /R112 161 0 R /R110 164 0 R /R116 169 0 R /R104 172 0 R /R114 175 0 R /R106 178 0 R /F2 181 0 R /R28 41 0 R /R38 45 0 R /R30 59 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R119 139 0 R /R122 182 0 R >> >> /MediaBox [ 0 0 612 792 ] /Annots [ 184 0 R 185 0 R 186 0 R 187 0 R 188 0 R 189 0 R 190 0 R ] >> endobj 6 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 191 0 R /Resources << /ExtGState << /R27 33 0 R >> /Font << /F2 192 0 R /R79 79 0 R /R73 83 0 R /R71 86 0 R /R77 90 0 R /R75 93 0 R /R127 145 0 R /R28 41 0 R /R125 148 0 R /F1 193 0 R /R85 98 0 R /R38 45 0 R /R81 105 0 R /R30 59 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 194 0 R 195 0 R 196 0 R 197 0 R 198 0 R 199 0 R 200 0 R 201 0 R 202 0 R 203 0 R 204 0 R 205 0 R ] >> endobj 7 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 206 0 R /Resources << /ExtGState << /R27 33 0 R >> /Font << /F2 207 0 R /R28 41 0 R /R172 208 0 R /R30 59 0 R /R38 45 0 R /R81 105 0 R /R34 53 0 R /R85 98 0 R /F1 212 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 213 0 R 214 0 R 215 0 R 216 0 R 217 0 R 218 0 R 219 0 R 220 0 R 221 0 R 222 0 R 223 0 R 224 0 R 225 0 R 226 0 R 227 0 R 228 0 R 229 0 R 230 0 R 231 0 R 232 0 R ] >> endobj 8 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 233 0 R /Resources << /XObject << /R202 234 0 R /R203 235 0 R /R200 236 0 R /R201 237 0 R /R206 238 0 R /R207 239 0 R /R204 240 0 R /R205 241 0 R /R220 242 0 R /R221 243 0 R /R222 244 0 R /R209 245 0 R /R190 246 0 R /R191 247 0 R /R196 248 0 R /R197 249 0 R /R194 250 0 R /R195 251 0 R /R189 252 0 R /R188 253 0 R /R208 254 0 R /R192 255 0 R /R210 256 0 R /R213 257 0 R /R212 258 0 R /R215 259 0 R /R214 260 0 R /R217 261 0 R /R216 262 0 R /R219 263 0 R /R218 264 0 R /R198 265 0 R /R199 266 0 R /R211 267 0 R /R193 268 0 R /R187 269 0 R >> /ExtGState << /R27 33 0 R >> /Font << /R79 79 0 R /R71 86 0 R /R223 270 0 R /R77 90 0 R /F2 273 0 R /R28 41 0 R /F1 274 0 R /R85 98 0 R /R38 45 0 R /R30 59 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R122 182 0 R >> >> /MediaBox [ 0 0 612 792 ] /Annots [ 275 0 R 276 0 R 277 0 R 278 0 R 279 0 R 280 0 R 281 0 R 282 0 R 283 0 R ] >> endobj 9 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 284 0 R /Resources << /XObject << /R243 285 0 R /R240 286 0 R >> /ExtGState << /R27 33 0 R >> /Font << /R108 156 0 R /R238 287 0 R /R110 164 0 R /R172 208 0 R /F2 291 0 R /R28 41 0 R /F1 292 0 R /R38 45 0 R /R30 59 0 R /R241 293 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] /ColorSpace << /R122 182 0 R >> >> /MediaBox [ 0 0 612 792 ] /Annots [ 297 0 R 298 0 R 299 0 R 300 0 R 301 0 R 302 0 R ] >> endobj 10 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 303 0 R /Resources << /XObject << /R266 304 0 R /R267 305 0 R /R268 306 0 R >> /ExtGState << /R27 33 0 R >> /Font << /F2 307 0 R /R28 41 0 R /F1 308 0 R /R85 98 0 R /R38 45 0 R /R81 105 0 R /R77 90 0 R /R30 59 0 R /R34 53 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 309 0 R 310 0 R 311 0 R 312 0 R 313 0 R 314 0 R 315 0 R 316 0 R 317 0 R 318 0 R 319 0 R 320 0 R 321 0 R 322 0 R 323 0 R ] >> endobj 11 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 324 0 R /Resources << /ExtGState << /R27 33 0 R >> /Font << /F2 325 0 R /R28 41 0 R /F1 326 0 R /R30 59 0 R /R38 45 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 327 0 R 328 0 R 329 0 R 330 0 R 331 0 R 332 0 R 333 0 R 334 0 R 335 0 R 336 0 R 337 0 R 338 0 R 339 0 R 340 0 R 341 0 R 342 0 R 343 0 R 344 0 R 345 0 R 346 0 R 347 0 R 348 0 R 349 0 R 350 0 R 351 0 R 352 0 R 353 0 R 354 0 R 355 0 R 356 0 R 357 0 R 358 0 R 359 0 R 360 0 R 361 0 R 362 0 R 363 0 R 364 0 R 365 0 R 366 0 R 367 0 R 368 0 R 369 0 R 370 0 R 371 0 R 372 0 R 373 0 R 374 0 R 375 0 R 376 0 R 377 0 R ] >> endobj 12 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 378 0 R /Resources << /ExtGState << /R27 33 0 R >> /Font << /F2 379 0 R /F1 380 0 R /R30 59 0 R /R38 45 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 381 0 R 382 0 R 383 0 R 384 0 R 385 0 R 386 0 R 387 0 R 388 0 R 389 0 R 390 0 R 391 0 R ] >> endobj 13 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 14 0 obj << /Length 20940 >> stream q q q 0.10000 0 0 0.10000 0 0 cm /R27 gs 0 g q 10 0 0 10 0 0 cm BT /R28 14.34620 Tf 1 0 0 1 191.92500 675.06700 Tm [ (Lip) -250.00800 (Reading) -249.99500 (Sentences) -249.99300 (in) -249.99300 (the) -250.00800 (W) 18.00480 (ild) ] TJ /R30 11.95520 Tf -114.45800 -37.85820 Td [ (Joon) -250.00800 (Son) -249.99300 (Chung) ] TJ /R32 7.97010 Tf 79.04690 4.33906 Td [ (1) -0.29866 ] TJ /R34 6.97380 Tf -79 -18.28590 Td (joon\100robots\056ox\056ac\056uk) Tj /R30 11.95520 Tf 120.82400 13.94690 Td [ (Andre) 24.98550 (w) -249.98700 (Senior) ] TJ /R32 7.97010 Tf 72.41210 4.33906 Td [ (2) -0.30019 ] TJ /R34 6.97380 Tf -81.95900 -18.28590 Td (andrewsenior\100google\056com) Tj /R30 11.95520 Tf 126.81500 13.94690 Td [ (Oriol) -249.99500 (V) 59.99160 (in) 14.99540 (yals) ] TJ /R32 7.97010 Tf 64.52230 4.33906 Td [ (2) -0.30019 ] TJ /R34 6.97380 Tf -67.55390 -18.28590 Td (vinyals\100google\056com) Tj /R30 11.95520 Tf 102.86200 13.94690 Td [ (Andre) 24.98340 (w) -249.98900 (Zisserman) ] TJ /R32 7.97010 Tf 91.00200 4.33906 Td [ (1) -0.30019 ] TJ /R36 7.97010 Tf 4.23438 0 Td [ (\073) -0.09802 ] TJ /R32 7.97010 Tf 2.35273 0 Td [ (2) -0.30019 ] TJ /R34 6.97380 Tf -84.08670 -18.28590 Td (az\100robots\056ox\056ac\056uk) Tj /R32 7.97010 Tf -310.79400 -15.58710 Td [ (1) -0.29866 ] TJ /R30 11.95520 Tf 7.72109 -4.33906 Td [ (Department) -250 (of) -250.01500 (Engineering) -249.99700 (Science\054) -249.99300 (Uni) 24.99570 (v) 14.98510 (ersity) -249.98900 (of) -250.01400 (Oxford) ] TJ /R32 7.97010 Tf 292.36200 4.33906 Td [ (2) -0.30019 ] TJ /R30 11.95520 Tf 7.72109 -4.33906 Td (DeepMind) Tj /R28 11.95520 Tf -279.95300 -41.04610 Td (Abstract) Tj /R38 9.96260 Tf -83.92770 -24.03090 Td [ (The) -346.00500 (goal) -347.01800 (of) -346.00800 (this) -347.01400 (work) -346.00100 (is) -346.98700 (to) -346.00800 (r) 37.01960 (eco) 9.99466 (gnise) -346.98100 (phr) 14.99500 (ases) -346.00800 (and) -346.98600 (sen\055) ] TJ -11.95510 -11.95590 Td [ (tences) -324.01900 (being) -324.98700 (spok) 10.01060 (en) -323.98400 (by) -325.00300 (a) -324.01900 (talking) -323.99200 (face) 10.00200 (\054) -343 (with) -325.01200 (or) -324.01200 (without) -324.99500 (the) ] TJ 11.95510 TL T* [ (audio\056) -381.00500 (Unlik) 10.01610 (e) -273.99800 (pr) 36.98520 (e) 15.01220 (vious) -274.01400 (works) -274.00700 (that) -272.98900 (have) -273.99100 (focussed) -274.00800 (on) -273.98400 (r) 37.01830 (eco) 9.99588 (g\055) ] TJ T* [ (nising) -297.01300 (a) -297.00400 (limited) -295.99700 (number) -296.99000 (of) -296.99700 (wor) 36.99870 (ds) -296.99800 (or) -296.99800 (phr) 14.99500 (ases\054) -308.01200 (we) -297 (tac) 20.00400 (kle) -297.01200 (lip) ] TJ T* [ (r) 37.01960 (eading) -246.99500 (as) -245.98700 (an) ] TJ /R30 9.96260 Tf 56.81290 0 Td [ (open\055w) 10.00810 (orld) ] TJ /R38 9.96260 Tf 48.28400 0 Td [ (pr) 44.98390 (oblem) -245.98200 (\226) -247.01300 (unconstr) 15.01220 (ained) -247 (natur) 15.00730 (al) ] TJ -105.09700 -11.95470 Td [ (langua) 9.99159 (g) 10.00320 (e) -250.00200 (sentences\054) -249.98300 (and) -249.98300 (in) -249.98500 (the) -249.99000 (wild) -250.00400 (videos\056) ] TJ 11.95510 -12.07730 Td [ (Our) -314.01200 (k) 9.99404 (e) 29.99070 (y) -313.01100 (contrib) 19.98810 (utions) -313.99100 (ar) 36.98520 (e\072) -436.99300 (\0501\051) -313.98800 (a) -313.98200 (\140W) 91.98220 (atc) 14.98520 (h\054) -328.99700 (Listen\054) -329.98700 (Attend) ] TJ -11.95510 -11.95470 Td [ (and) -251.00200 (Spell\047) -251 (\050WLAS\051) -250.99900 (network) -250.98900 (that) -250.99200 (learns) -250.99200 (to) -251.00500 (tr) 14.99010 (anscribe) -250.99200 (videos) ] TJ T* [ (of) -344.98900 (mouth) -345.99600 (motion) -345.00900 (to) -346.00800 (c) 15.01220 (har) 14.99380 (acter) 10.00810 (s\073) -392.99100 (\0502\051) -345.00200 (a) -344.99600 (curriculum) -345.98400 (learning) ] TJ T* [ (str) 14.99750 (ate) 40.00130 (gy) -230 (to) -231.00800 (acceler) 15.01830 (ate) -229.99300 (tr) 14.99140 (aining) -230.00600 (and) -229.98600 (to) -231.00800 (r) 37.01960 (educe) -230.00300 (o) 10.00320 <766572027474696e673b> -236.98700 (\0503\051) ] TJ T* [ (a) -307.00300 (\140Lip) -306.00400 (Reading) -306.98100 (Sentences\047) -307 (\050LRS\051) -306.99000 (dataset) -305.99600 (for) -306.99000 (visual) -306.98100 (speec) 15.01470 (h) ] TJ 11.95590 TL T* [ (r) 37.01960 (eco) 9.99466 (gnition\054) -406.98600 (consisting) -376.00600 (of) -376.00300 (o) 10.00320 (ver) -376.01800 (100\054000) -375.98400 (natur) 15.00730 (al) -376.00300 (sentences) ] TJ 11.95510 TL T* [ (fr) 44.98640 (om) -249.99000 (British) -249.99400 (tele) 14.98400 (vision\056) ] TJ 11.95510 -12.07580 Td [ (The) -316.01000 (WLAS) -316.00100 (model) -316.01000 (tr) 14.99140 (ained) -316.00700 (on) -316.01600 (the) -316.99900 (LRS) -316.00400 (dataset) -315.99400 (surpasses) ] TJ -11.95510 -11.95510 Td [ (the) -271.98600 (performance) -272.98400 (of) -271.98200 (all) -272.99400 (pr) 36.98520 (e) 15.01220 (vious) -272.01500 (work) -272.01300 (on) -273.00300 (standar) 37.01340 (d) -271.98900 (lip) -272.99400 (r) 37.01830 (ead\055) ] TJ T* [ (ing) -247.98100 (benc) 15.01830 (hmark) -247.00300 (datasets\054) -249.01700 (often) -246.99800 (by) -247.99700 (a) -247.99300 <7369676e690263616e74> -247.00800 (mar) 36.98890 (gin\056) -310.00700 (This) ] TJ 11.95590 TL T* [ (lip) -284.99200 (r) 37.01960 (eading) -284.00800 (performance) -284.98200 (beats) -284.99300 (a) -285.00600 (pr) 44.98390 (ofessional) -283.98200 (lip) -284.99200 (r) 37.01830 (eader) -285.00400 (on) ] TJ 11.95510 TL T* [ (videos) -225.98700 (fr) 44.98640 (om) -225.99400 (BBC) -226.02000 (tele) 14.98400 (vision\054) -230.99200 (and) -225.98700 (we) -225.99300 (also) -226.01900 (demonstr) 15.00980 (ate) -225.99400 (that) -226.01600 (if) ] TJ T* [ (audio) -232.01300 (is) -233.00600 (available) 9.99220 (\054) -235.01700 (then) -233.00800 (visual) -232.01400 (information) -232.01500 (helps) -231.98100 (to) -233.00800 (impr) 44.99490 (o) 10.00320 (ve) ] TJ T* [ (speec) 15.01530 (h) -249.99300 (r) 37.01960 (eco) 9.99466 (gnition) -249.99500 (performance) 15.00730 (\056) ] TJ /R28 11.95520 Tf 36.22930 TL T* [ (1\056) -249.99000 (Intr) 18.01460 (oduction) ] TJ /R30 9.96260 Tf 11.95510 -14.06870 Td [ (Lip) -330.00700 (reading\054) -350.00800 (the) -330.01600 (ability) -329.00900 (to) -330.01100 (recognize) -329.98200 (what) -330.01400 (is) -330.01100 (being) -330.00600 (said) ] TJ -11.95510 -11.95510 Td [ (from) -358.00800 (visual) -359.01200 (information) -358.01500 (alone\054) -386.00300 (is) -358.00500 (an) -358.99800 (impressi) 25.01300 (v) 14.98280 (e) -357.98400 (skill\054) -386.01900 (and) ] TJ 11.95510 TL T* [ (v) 14.98280 (ery) -349.00300 (challenging) -349.98700 (for) -349.00200 (a) -349.00500 (no) 14.98770 (vice\056) -608.01000 (It) -348.99700 (is) -348.98700 (inherently) -349.99100 (ambiguous) ] TJ 11.95590 TL T* [ (at) -255.01300 (the) -255.00800 (w) 10.00140 (ord) -255.01000 (le) 25.01790 (v) 14.98280 (el) -255.01400 (due) -255.01000 (to) -255.00400 (homophemes) -255.00900 (\226) -255.01100 (dif) 24.98600 (ferent) -255.01900 (characters) ] TJ 11.95510 TL T* [ (that) -201.99000 (produce) -201.01700 (e) 15.01220 (xactly) -201.99000 (the) -201.01800 (same) -201.99600 (lip) -201.00600 (sequence) -202 (\050) ] TJ /R38 9.96260 Tf 172.71100 0 Td [ (e) 15.01220 (\056g) 14.98520 (\056) ] TJ /R30 9.96260 Tf 16.09610 0 Td [ (\140p\047) -202.00800 (and) -201.02000 (\140b\047\051\056) ] TJ -188.80700 -11.95510 Td [ (Ho) 24.98420 (we) 25.01420 (v) 14.98280 (er) 39.98600 (\054) -259.99300 (such) -257.98400 (ambiguities) -259 (can) -258 (be) -257.99500 (resolv) 14.99260 (ed) -257.99600 (to) -257.98400 (an) -259.01600 (e) 15.01220 (xtent) -258.01600 (us\055) ] TJ T* [ (ing) -300.01100 (the) -299.98100 (conte) 14.98650 (xt) -300.99600 (of) -299.98800 (neighboring) -299.98700 (w) 10 (ords) -300.01600 (in) -300.99600 (a) -299.99400 (sentence\054) -313 (and\057or) ] TJ T* [ (a) -250.00200 (language) -249.98300 (model\056) ] TJ 11.95510 -12.07700 Td [ (A) -374.99300 (machine) -376 (that) -374.98200 (can) -375 (lip) -375.99600 (read) -375.00400 (opens) -375.01800 (up) -376.00600 (a) -375.00100 (host) -375.01300 (of) -376.01300 (appli\055) ] TJ -11.95510 -11.95470 Td [ (cations\072) -456.01300 (\140dictating\047) -323.01600 (instructi) 1.01699 (on) -1.00964 (s) -322.01900 (or) -323.00400 (messages) -322.98300 (to) -322.99300 (a) -323.01000 (phone) -322.99000 (in) ] TJ T* [ (a) -331.98800 (nois) 1.00351 (y) -332.01800 (en) 39.99390 (vironment\073) -372.00900 (transcribing) -331.99100 (and) -330.99700 (re\055dubbing) -331.99600 (archi) 25.01540 (v) 24.98110 (al) ] TJ T* [ (silent) -488.00800 <026c6d733b> -606.00900 (resolving) -488.00300 (multi\055talk) 10.00940 (er) -486.98500 (simultaneous) -488.01900 (speech\073) ] TJ T* [ (and\054) -725.00800 (impro) 15.00420 (ving) -629.01000 (the) -630.00500 (performance) -628.99300 (of) -630.01000 (automated) -629.98300 (speech) ] TJ 11.95590 TL T* [ (recogition) -250.00700 (in) -249.98500 (general\056) ] TJ 11.95510 -12.07620 Td [ (That) -304.99600 (such) -304.99600 (automation) -305.00200 (is) -304.99400 (no) 24.98600 (w) -305.00500 (possible) -305.01000 (is) -304.99600 (due) -305.00300 (to) -304.99600 (tw) 10.00810 (o) -305.00300 (de\055) ] TJ -11.95510 -11.95510 Td [ (v) 14.98280 (elopments) -464.98100 (that) -465.00600 (are) -466.01300 (well) -465.00800 (kno) 24.99090 (wn) -465.01300 (across) -465.01600 (computer) -466.01800 (vision) ] TJ 258.75000 481.29000 Td [ (tasks\072) -401.00400 (the) -295.98200 (use) -295.00200 (of) -295.98700 (deep) -295.00700 (neural) -295.98500 (netw) 10.00810 (ork) -295.00500 (models) -296 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 498.27100 562.29000 Tm (22) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 508.23300 562.29000 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 513.66300 562.29000 Tm (33) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 523.62500 562.29000 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 529.06500 562.29000 Tm (35) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 539.02700 562.29000 Tm (\135\073) Tj -230.16500 -11.95510 Td [ (and\054) -219.01500 (the) -209.99700 (a) 19.99180 (v) 24.98110 (ailability) -211.01900 (of) -210.98200 (a) -210.98900 (lar) 17.99700 (ge) -210.98400 (scale) -210.00400 (dataset) -211.00900 (for) -210.98400 (training) -210.99400 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 529.34500 550.33500 Tm (31) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 539.30700 550.33500 Tm (\135\056) Tj -230.44500 -11.95510 Td [ (In) -327.00200 (this) -328 (case) -327.01200 (the) -328.01600 (model) -327.01900 (is) -328.01100 (based) -326.99400 (on) -328.01400 (the) -326.99400 (recent) -327.98900 (sequence\055to\055) ] TJ 11.95590 TL T* [ (sequence) -281.98700 (\050encoder) 19.99180 (\055decoder) -281.99700 (wit) 1 (h) -281.98700 (attention\051) -282.00200 (translater) -282.01700 (archi\055) ] TJ 11.95510 TL T* [ (tectures) -406.98100 (that) -407.01500 (ha) 19.99670 (v) 14.98280 (e) -408.01500 (been) -406.98800 (de) 25.01540 (v) 14.98280 (eloped) -407.01500 (for) -406.99100 (speech) -408.00500 (recognition) ] TJ T* [ (and) -297.98500 (machine) -298.01200 (translation) -297.99200 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 411.28800 502.51400 Tm (3) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 416.26900 502.51400 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 421.72800 502.51400 Tm (5) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 426.70900 502.51400 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 432.15900 502.51400 Tm (15) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 442.12100 502.51400 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 447.58100 502.51400 Tm (16) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 457.54300 502.51400 Tm (\054) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 463.00300 502.51400 Tm (34) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 472.96500 502.51400 Tm [ (\135\056) -454.00300 (The) -297.98500 (dataset) -298.01400 (de\055) ] TJ -164.10300 -11.95510 Td [ (v) 14.98280 (eloped) -207.99000 (in) -207.99200 (this) -207 (paper) -208.00700 (is) -207.99200 (based) -207.99500 (on) -207.99500 (thousands) -206.98500 (of) -208.00500 (hours) -207.98500 (of) -208.00500 (BBC) ] TJ 11.95470 TL T* [ (tele) 24.98110 (vision) -343.00700 (broadcasts) -342.00400 (that) -342.98700 (ha) 19.99670 (v) 14.98280 (e) -343.00700 (talking) -341.98700 (f) 9.99343 (aces) -343.00700 (together) -342.98200 (with) ] TJ T* [ (subtitles) -249.99000 (of) -249.99500 (what) -249.98800 (is) -249.98500 (being) -250.02000 (said\056) ] TJ 11.95510 -12.82300 Td [ (W) 79.98660 (e) -226.00600 (also) -225.98700 (in) 40.00560 (v) 14.98280 (estig) 5.00162 (ate) -226.98400 (ho) 24.98600 (w) -225.99700 (lip) -225.98200 (reading) -225.99200 (can) -226.00600 (contrib) 20.01630 (ute) -227.01100 (to) ] TJ /R38 9.96260 Tf 211.01500 0 Td (au\055) Tj -222.97000 -11.95510 Td (dio) Tj /R30 9.96260 Tf 15.48400 0 Td [ (based) -275.98600 (speech) -277.01000 (recogniti) 1.00473 (o) -1.01454 (n\056) -388.00400 (There) -277.01700 (is) -276.01800 (a) -275.99800 (lar) 17.99700 (ge) -275.99300 (literature) -276.01800 (on) ] TJ -15.48400 -11.95470 Td [ (this) -195.00200 (contrib) 20.01870 (ution\054) -206.01200 (particularly) -193.98200 (in) -195.01600 (noisy) -194.99700 (en) 39.99330 (vironments\054) -205.99500 (as) -194.98700 (well) ] TJ 11.95630 TL T* [ (as) -319.00300 (the) -318.99800 (con) 39.99820 (v) 14.98280 (erse) -319.01000 (where) -319.01000 (some) -318.98600 (deri) 24.98110 (v) 14.98280 (ed) -319.00600 (measure) -320.02000 (of) -319.00600 (audio) -318.98600 (can) ] TJ 11.95510 TL T* [ (contrib) 20.01870 (ute) -265.98700 (to) -265.00200 (lip) -266.01400 (reading) -265.98500 (for) -265.01500 (the) -265.98500 (deaf) -266 (or) -265.01500 (hard) -265.99500 (of) -265.99500 (hearing\056) -356.98500 (T) 79.99160 (o) ] TJ T* [ (in) 40.00560 (v) 14.98280 (estig) 5.00162 (ate) -317.00800 (this) -316.98100 (aspect) -317.98100 (we) -317.00600 (train) -316.99600 (a) -317.01100 (model) -316.98100 (to) -318.01600 (recognize) -317.00600 (char) 19.99180 (\055) ] TJ T* [ (acters) -255.01900 (from) -255.98700 (both) -254.99400 (audio) -256.01900 (a) 1.01454 (nd) -255.98700 (visual) -254.99200 (input\054) -256.98200 (and) -255.01100 (then) -255.98200 (systemat\055) ] TJ T* [ (ically) -280.99500 (disturb) -280 (the) -281.00500 (audio) -280.99500 (channel) -279.98300 (or) -281.01200 (remo) 14.98770 (v) 14.98280 (e) -279.99700 (the) -281.00200 (visual) -280.98700 (chan\055) ] TJ (nel\056) ' 11.95510 -12.82300 Td [ (Our) -321.98100 (model) -323.02000 (\050Section) ] TJ ET Q 1 0 0 rg q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 404.05500 345.36200 Tm (2) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 409.03600 345.36200 Tm [ (\051) -321.99000 (outputs) -321.98500 (at) -323 (the) -322.01500 (character) -322 (le) 25.02030 (v) 14.98280 (el\054) -340.99700 (is) ] TJ -100.17400 -11.95470 Td [ (able) -386.01200 (to) -384.98200 (learn) -386.01400 (a) -386.01900 (language) -386 (model\054) -418.99900 (and) -386.00900 (has) -386.00900 (a) -385 (no) 14.98770 (v) 14.98280 (el) -386.00900 (dual) -386 (at\055) ] TJ 11.95510 TL T* [ (tention) -299.01600 (mechanism) -299.99400 (that) -298.99400 (can) -299.01400 (operate) -299.98900 (o) 14.98280 (v) 14.98280 (er) -299.01900 (visual) -298.98400 (input) -299.99900 (only) 64.99410 (\054) ] TJ 11.95590 TL T* [ (audio) -316.98900 (input) -317.99600 (only) 64.99160 (\054) -333.98100 (or) -317.00300 (both\056) -512.98800 (W) 79.98660 (e) -317.01100 (sho) 24.99090 (w) -317.98100 (\050Secti) 0.99493 (on) ] TJ ET Q 1 0 0 rg q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 493.96700 309.49600 Tm (3) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 498.94800 309.49600 Tm [ (\051) -317.01100 (that) -318.01100 (train\055) ] TJ -190.08600 -11.95510 Td [ (ing) -385.99700 (can) -384.99700 (be) -386.01400 (accelerated) -385.99000 (by) -384.98500 (a) -386.01900 (form) -384.98500 (of) -386.01400 (curriculum) -385.99000 (learning\056) ] TJ 11.95510 TL T* [ (W) 79.98660 (e) -371.00200 (also) -370.00200 (describe) -370.99000 (\050Section) ] TJ ET Q 1 0 0 rg q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 419.10800 285.58600 Tm (4) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 424.08900 285.58600 Tm [ (\051) -371.00200 (the) -370.98700 (generation) -369.99700 (and) -370.99200 (statistics) ] TJ -115.22700 -11.95510 Td [ (of) -285.00900 (a) -285.99600 (ne) 25.01540 (w) -285.00900 (lar) 17.99700 (ge) -285.01100 (scale) -285.99100 (Lip) -284.99400 (Reading) -284.98700 (Sentences) -284.99200 (\050LRS\051) -286.00600 (dataset\054) ] TJ 11.95470 TL T* [ (based) -271.00600 (on) -269.98400 (BBC) -271.01400 (broadcasts) -270.01400 (containing) -271.01400 (talking) -270 (f) 9.99343 (aces) -270.97900 (together) ] TJ T* [ (with) -270.01100 (subtitles) -269.98700 (of) -271.01100 (what) -269.98400 (is) -269.98200 (said\056) -370.98200 (The) -269.98900 (broadcasts) -270.01400 (contain) -270.99400 (f) 9.99343 (aces) ] TJ 11.95630 TL T* [ (\140in) -257.00600 (the) -255.98900 (wild\047) -256.99600 (with) -256.99400 (a) -256.00100 <7369676e690263616e74> -257.00600 (v) 24.98110 (ariety) -257.00600 (of) -255.99600 (pose\054) -259 (e) 15.01220 (xpressions\054) ] TJ 11.95470 TL T* [ (lighting\054) -249.98500 (backgrounds\054) -250.00500 (and) -249.99300 (ethnic) -249.98300 (origin\056) ] TJ 11.95510 -12.82230 Td [ (The) -229.01600 (performance) -228.98100 (of) -229.01800 (the) -229.01100 (model) -228.99600 (is) -229.00600 (assessed) -229.00600 (on) -229.01100 (a) -228.98600 (test) -229.00600 (set) -229.01100 (of) ] TJ -11.95510 -11.95590 Td [ (the) -303 (LRS) -303.99600 (dataset\054) -316.99100 (as) -303.00600 (well) -304.01500 (as) -303.00800 (on) -304.01800 (public) -302.98400 (benchmarks) -304.00800 (datasets) ] TJ T* [ (for) -428.00900 (lip) -427.98700 (reading) -427.99700 (including) -428.98200 (LR) 54.99080 (W) -428.00700 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 444.75200 189.07700 Tm (9) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 449.73300 189.07700 Tm [ (\135) -428.01200 (and) -428.00200 (GRID) -427.99200 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 507.89500 189.07700 Tm (11) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 517.85700 189.07700 Tm [ (\135\056) -844.99600 (W) 79.98660 (e) ] TJ -208.99500 -11.95510 Td (demonstrate) Tj /R38 9.96260 Tf 50.90510 0 Td [ (open) -221.99200 (world) ] TJ /R30 9.96260 Tf 47.03590 0 Td [ (\050unconstrained) -222.01200 (sentences\051) -221.01200 (lip) -221.98300 (read\055) ] TJ -97.94100 -11.95510 Td [ (ing) -342.98400 (on) -344.01100 (the) -342.99400 (LRS) -343.00900 (dataset\054) -366.98300 (and) -342.99700 (in) -342.99200 (all) -344.01100 (cases) -343.00200 (on) -342.99200 (public) -343.99700 (bench\055) ] TJ 11.95470 TL T* [ (marks) -249.98300 (the) -249.99000 (performance) -249.99700 (e) 15.01220 (xceeds) -249.99300 (that) -249.98300 (of) -249.99700 (prior) -249.98800 (w) 10.00320 (ork\056) ] TJ /R28 10.95890 Tf 22.52810 TL T* [ (1\0561\056) -250.00400 (Related) -249.99300 (w) 10 (orks) ] TJ /R28 9.96260 Tf 13.81910 TL T* [ (Lip) -291.99500 (r) 17.99210 (eading) 15.01960 (\056) ] TJ /R30 9.96260 Tf 57.03980 0 Td [ (There) -291.99300 (is) -292.99800 (a) -291.99500 (lar) 17.99700 (ge) -291.99000 (body) -292.99000 (of) -291.99000 (w) 10.00320 (ork) -293.00500 (on) -291.98100 (lip) -292.01000 (reading) ] TJ -57.03980 -11.95470 Td [ (using) -294.98000 (pre\055deep) -293.99500 (learning) -294.99200 (methods\056) -442.98500 (These) -295.00500 (methods) -293.99500 (are) -294.98000 (thor) 20.01140 (\055) ] TJ 11.95510 TL T* [ (oughly) -329.99100 (re) 25.00810 (vie) 24.98360 (wed) -331.00100 (in) -330.01100 (\133) ] TJ ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 393.49500 92.95550 Tm (40) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R30 9.96260 Tf 1 0 0 1 403.45700 92.95550 Tm [ (\135\054) -351 (and) -330.01800 (we) -329.98400 (will) -331.01800 (not) -330.00400 (repeat) -329.98900 (this) -331.01800 (here\056) ] TJ -94.59490 -11.95510 Td [ (A) -331.00100 (number) -330.98400 (of) -329.98200 (papers) -330.99900 (ha) 19.99670 (v) 14.98280 (e) -331.00800 (used) -330.99400 (Con) 40.01780 (v) 20.00160 (olutional) -330.99900 (Neural) -330.99900 (Net\055) ] TJ -13.74100 -29.88790 Td (1) Tj ET Q Q Q q q 1 1 1 rg /a0 gs 48.40600 786.42200 515.18800 -52.69900 re f q /s5 gs /x6 Do Q q /s7 gs /x8 Do Q q /s9 gs /x10 Do Q q /s11 gs /x12 Do Q Q Q Q q 1 0 0 1 0 0 cm BT /F1 12 Tf 14.40000 TL ET 1 1 1 rg n 270 47 72 14 re f* 0.50000 0.50000 0.50000 rg BT /F2 9 Tf 10.80000 TL ET BT 1 0 0 1 297 50 Tm (6447) Tj T* ET Q endstream endobj 15 0 obj << /Filter /FlateDecode /Resources << /ExtGState << /a0 << /CA 1 /ca 1 >> >> /XObject << /x18 16 0 R >> >> /Length 28 /Group << /Type /Group /S /Transparency /CS /DeviceRGB /I true >> /BBox [ 78 746 96 765 ] /Type /XObject /Subtype /Form >> stream x+O4PH/VЯ0Pp 0 endstream endobj 16 0 obj << /Filter /FlateDecode /Resources 17 0 R /Length 107 /Type /XObject /BBox [ 78 746 96 765 ] /Subtype /Form >> stream xe AC̬wʠ =p,?]%+H-
Jc "82w8VSnGW;"
endstream
endobj
17 0 obj
<<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
>>
endobj
18 0 obj
<<
/Filter /FlateDecode
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x15 19 0 R
>>
>>
/Length 28
/Group <<
/Type /Group
/S /Transparency
/CS /DeviceRGB
/I true
>>
/BBox [ 67 752 84 775 ]
/Type /XObject
/Subtype /Form
>>
stream
x+O4PH/VЯ04Up
0
endstream
endobj
19 0 obj
<<
/Filter /FlateDecode
/Resources 20 0 R
/Length 228
/Type /XObject
/BBox [ 67 752 84 775 ]
/Subtype /Form
>>
stream
xeQKn!s ?FPav6R٪TS.
b];15YyR
{7QL.\:Rv/x9l+L7h%1!}i/AI(kz"U&,YO![R hg{3}4/GyYF:!w}Gn+'xJcO9i뽼_-:`
endstream
endobj
20 0 obj
<<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
>>
endobj
21 0 obj
<<
/Filter /FlateDecode
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x24 22 0 R
>>
>>
/Length 28
/Group <<
/Type /Group
/S /Transparency
/CS /DeviceRGB
/I true
>>
/BBox [ 132 751 480 772 ]
/Type /XObject
/Subtype /Form
>>
stream
x+O4PH/VЯ02Qp
0
endstream
endobj
22 0 obj
<<
/Filter /FlateDecode
/Resources 23 0 R
/Length 53223
/Type /XObject
/BBox [ 132 751 480 772 ]
/Subtype /Form
>>
stream
xtI:6%Q㨈?7rA= u%6 ?Y(WbWo{B>9
x`Znϳ|8{3?0x*z ǃ|,@:w>`c|*ϻⳅKO3`g
:_|}}><.6`Z{{3]#<_o"~:ͺgk7/Ұ@|K yp ]03ʷCmş8˽Y?>(3!Bwqs.Z8,~~=rMT̩y+/*w: uBZ_`ߵp`%M?ɝ1ɳw=vDۉy&xb4Q>d@ sg~lA