Your browser does not support the video tag. Please use IE9+ or Google Chrome.
投影片 1 (Lab531, 1:05:50)
 
 
 
  • 1. Slide 1
  • 2. Possible Candidatesphrases, words, syllables, phonemes.....Phonemethe minimum units of speech sound in a language which can serve to distinguish one word from the othere.g. b at / p at , b a d / b e dphone : a phoneme’s acoustic realization the same phoneme may have many different realizations e.g. sa t / me t erCoarticulation and Context Dependencycontext: right/left neighboring unitscoarticulation: sound production changed because of the neighboring unitsright-context-dependent (RCD)/left-context-dependent (LCD)/ bothintraword/interword context dependencyFor Mandarin Chinesecharacter/syllable mapping relationsyllable: Initial (聲母) / Final (韻母) / tone (聲調)
  • 3. Primary Considerationsaccuracy: accurately representing the acoustic realizationstrainability: feasible to obtain enough data to estimate the model parametersgeneralizability: any new word can be derived from a predefined unit inventoryExampleswords: accurate if enough data available, trainable for small vocabulary, NOT generalizablephoneme : trainable, generalizable difficult to be accurate due to context dependencysyllable: 50 in Japanese, 1300 in Mandarin Chinese, over 30000 in English Triphonea phoneme model taking into consideration both left and right neighboring phonemes(60)3→ 216,000very good generalizability, balance between accuracy/ trainability by parameter-sharing techniques
  • 4. Possible Candidatesphrases, words, syllables, phonemes.....Phonemethe minimum units of speech sound in a language which can serve to distinguish one word from the othere.g. b at / p at , b a d / b e dphone : a phoneme’s acoustic realization the same phoneme may have many different realizations e.g. sa t / me t erCoarticulation and Context Dependencycontext: right/left neighboring unitscoarticulation: sound production changed because of the neighboring unitsright-context-dependent (RCD)/left-context-dependent (LCD)/ bothintraword/interword context dependencyFor Mandarin Chinesecharacter/syllable mapping relationsyllable: Initial (聲母) / Final (韻母) / tone (聲調)
  • 5. Primary Considerationsaccuracy: accurately representing the acoustic realizationstrainability: feasible to obtain enough data to estimate the model parametersgeneralizability: any new word can be derived from a predefined unit inventoryExampleswords: accurate if enough data available, trainable for small vocabulary, NOT generalizablephoneme : trainable, generalizable difficult to be accurate due to context dependencysyllable: 50 in Japanese, 1300 in Mandarin Chinese, over 30000 in English Triphonea phoneme model taking into consideration both left and right neighboring phonemes(60)3→ 216,000very good generalizability, balance between accuracy/ trainability by parameter-sharing techniques
  • 6. Possible Candidatesphrases, words, syllables, phonemes.....Phonemethe minimum units of speech sound in a language which can serve to distinguish one word from the othere.g. b at / p at , b a d / b e dphone : a phoneme’s acoustic realization the same phoneme may have many different realizations e.g. sa t / me t erCoarticulation and Context Dependencycontext: right/left neighboring unitscoarticulation: sound production changed because of the neighboring unitsright-context-dependent (RCD)/left-context-dependent (LCD)/ bothintraword/interword context dependencyFor Mandarin Chinesecharacter/syllable mapping relationsyllable: Initial (聲母) / Final (韻母) / tone (聲調)
  • 7. Primary Considerationsaccuracy: accurately representing the acoustic realizationstrainability: feasible to obtain enough data to estimate the model parametersgeneralizability: any new word can be derived from a predefined unit inventoryExampleswords: accurate if enough data available, trainable for small vocabulary, NOT generalizablephoneme : trainable, generalizable difficult to be accurate due to context dependencysyllable: 50 in Japanese, 1300 in Mandarin Chinese, over 30000 in English Triphonea phoneme model taking into consideration both left and right neighboring phonemes(60)3→ 216,000very good generalizability, balance between accuracy/ trainability by parameter-sharing techniques
  • 8. Possible Candidatesphrases, words, syllables, phonemes.....Phonemethe minimum units of speech sound in a language which can serve to distinguish one word from the othere.g. b at / p at , b a d / b e dphone : a phoneme’s acoustic realization the same phoneme may have many different realizations e.g. sa t / me t erCoarticulation and Context Dependencycontext: right/left neighboring unitscoarticulation: sound production changed because of the neighboring unitsright-context-dependent (RCD)/left-context-dependent (LCD)/ bothintraword/interword context dependencyFor Mandarin Chinesecharacter/syllable mapping relationsyllable: Initial (聲母) / Final (韻母) / tone (聲調)
  • 9. Primary Considerationsaccuracy: accurately representing the acoustic realizationstrainability: feasible to obtain enough data to estimate the model parametersgeneralizability: any new word can be derived from a predefined unit inventoryExampleswords: accurate if enough data available, trainable for small vocabulary, NOT generalizablephoneme : trainable, generalizable difficult to be accurate due to context dependencysyllable: 50 in Japanese, 1300 in Mandarin Chinese, over 30000 in English Triphonea phoneme model taking into consideration both left and right neighboring phonemes(60)3→ 216,000very good generalizability, balance between accuracy/ trainability by parameter-sharing techniques
  • 10. Sharing of Parameters and Training Data for Triphones
  • 11. Primary Considerationsaccuracy: accurately representing the acoustic realizationstrainability: feasible to obtain enough data to estimate the model parametersgeneralizability: any new word can be derived from a predefined unit inventoryExampleswords: accurate if enough data available, trainable for small vocabulary, NOT generalizablephoneme : trainable, generalizable difficult to be accurate due to context dependencysyllable: 50 in Japanese, 1300 in Mandarin Chinese, over 30000 in English Triphonea phoneme model taking into consideration both left and right neighboring phonemes(60)3→ 216,000very good generalizability, balance between accuracy/ trainability by parameter-sharing techniques
  • 12. Sharing of Parameters and Training Data for Triphones
  • 13. Primary Considerationsaccuracy: accurately representing the acoustic realizationstrainability: feasible to obtain enough data to estimate the model parametersgeneralizability: any new word can be derived from a predefined unit inventoryExampleswords: accurate if enough data available, trainable for small vocabulary, NOT generalizablephoneme : trainable, generalizable difficult to be accurate due to context dependencysyllable: 50 in Japanese, 1300 in Mandarin Chinese, over 30000 in English Triphonea phoneme model taking into consideration both left and right neighboring phonemes(60)3→ 216,000very good generalizability, balance between accuracy/ trainability by parameter-sharing techniques
  • 14. Sharing of Parameters and Training Data for Triphones
  • 15. Primary Considerationsaccuracy: accurately representing the acoustic realizationstrainability: feasible to obtain enough data to estimate the model parametersgeneralizability: any new word can be derived from a predefined unit inventoryExampleswords: accurate if enough data available, trainable for small vocabulary, NOT generalizablephoneme : trainable, generalizable difficult to be accurate due to context dependencysyllable: 50 in Japanese, 1300 in Mandarin Chinese, over 30000 in English Triphonea phoneme model taking into consideration both left and right neighboring phonemes(60)3→ 216,000very good generalizability, balance between accuracy/ trainability by parameter-sharing techniques
  • 16. Sharing of Parameters and Training Data for Triphones
  • 17. Some Fundamentals in Information Theory
  • 18. Sharing of Parameters and Training Data for Triphones
  • 19. Primary Considerationsaccuracy: accurately representing the acoustic realizationstrainability: feasible to obtain enough data to estimate the model parametersgeneralizability: any new word can be derived from a predefined unit inventoryExampleswords: accurate if enough data available, trainable for small vocabulary, NOT generalizablephoneme : trainable, generalizable difficult to be accurate due to context dependencysyllable: 50 in Japanese, 1300 in Mandarin Chinese, over 30000 in English Triphonea phoneme model taking into consideration both left and right neighboring phonemes(60)3→ 216,000very good generalizability, balance between accuracy/ trainability by parameter-sharing techniques
  • 20. Sharing of Parameters and Training Data for Triphones
  • 21. Some Fundamentals in Information Theory
  • 22. Fundamentals in Information Theory
  • 23. Some Fundamentals in Information Theory
  • 24. Fundamentals in Information Theory
  • 25. Fundamentals in Information Theory
  • 26. Fundamentals in Information Theory
  • 27. Some Fundamentals in Information Theory
  • 28. Classification and Regression Trees (CART)
  • 29. Goal
  • 30. Splitting Criteria for the Decision Tree
  • 31. Training Triphone Models with Decision Trees
  • 32. Training Tri-phone Models with Decision Trees
  • 33. Some Fundamentals in Information Theory
  • 34. Fundamentals in Information Theory
  • 35. Some Fundamentals in Information Theory
  • 36. Fundamentals in Information Theory
  • 37. Some Fundamentals in Information Theory
  • 38. Fundamentals in Information Theory
  • 39. Some Fundamentals in Information Theory
  • 40. Fundamentals in Information Theory
  • 41. Some Fundamentals in Information Theory
  • 42. Fundamentals in Information Theory
  • 43. Some Fundamentals in Information Theory
  • 44. Fundamentals in Information Theory
  • 45. Some Fundamentals in Information Theory
  • 46. Fundamentals in Information Theory
  • 47. Some Fundamentals in Information Theory
  • 48. Fundamentals in Information Theory
  • 49. Some Fundamentals in Information Theory
  • 50. Fundamentals in Information Theory
  • 51. Some Fundamentals in Information Theory
  • 52. Fundamentals in Information Theory
  • 53. Some Fundamentals in Information Theory
  • 54. Fundamentals in Information Theory
  • 55. Some Fundamentals in Information Theory
  • 56. Some Fundamentals in Information Theory
  • 57. Fundamentals in Information Theory
  • 58. Some Fundamentals in Information Theory
  • 59. Fundamentals in Information Theory
  • 60. Some Fundamentals in Information Theory
  • 61. Fundamentals in Information Theory
  • 62. Some Fundamentals in Information Theory
  • 63. Fundamentals in Information Theory
  • 64. Some Fundamentals in Information Theory
  • 65. Fundamentals in Information Theory
  • 66. Some Fundamentals in Information Theory
  • 67. Fundamentals in Information Theory
  • 68. Some Fundamentals in Information Theory
  • 69. Fundamentals in Information Theory
  • 70. Some Fundamentals in Information Theory
  • 71. Fundamentals in Information Theory
  • 72. Some Fundamentals in Information Theory
  • 73. Fundamentals in Information Theory
  • 74. Some Fundamentals in Information Theory
  • 75. Fundamentals in Information Theory
  • 76. Some Fundamentals in Information Theory
  • 77. Fundamentals in Information Theory
  • 78. Some Fundamentals in Information Theory
  • 79. Fundamentals in Information Theory
  • 80. Fundamentals in Information Theory
  • 81. Fundamentals in Information Theory
  • 82. Fundamentals in Information Theory
  • 83. Fundamentals in Information Theory
  • 84. Fundamentals in Information Theory
  • 85. Fundamentals in Information Theory
  • 86. Fundamentals in Information Theory
  • 87. Fundamentals in Information Theory
  • 88. Fundamentals in Information Theory
  • 89. Fundamentals in Information Theory
  • 90. Fundamentals in Information Theory
  • 91. Fundamentals in Information Theory
  • 92. Some Fundamentals in Information Theory
  • 93. Fundamentals in Information Theory
  • 94. Fundamentals in Information Theory
  • 95. Fundamentals in Information Theory
  • 96. Fundamentals in Information Theory
  • 97. Fundamentals in Information Theory
  • 98. Fundamentals in Information Theory
  • 99. Fundamentals in Information Theory
  • 100. Fundamentals in Information Theory
  • 101. Fundamentals in Information Theory
  • 102. Fundamentals in Information Theory
  • 103. Fundamentals in Information Theory
  • 104. Some Fundamentals in Information Theory
  • 105. Fundamentals in Information Theory
  • 106. Some Fundamentals in Information Theory
  • 107. Fundamentals in Information Theory
  • 108. Some Fundamentals in Information Theory
  • 109. Fundamentals in Information Theory
  • 110. Fundamentals in Information Theory
  • 111. Fundamentals in Information Theory
  • 112. Some Fundamentals in Information Theory
  • 113. Fundamentals in Information Theory
  • 114. Fundamentals in Information Theory
  • 115. Fundamentals in Information Theory
  • 116. Some Fundamentals in Information Theory
  • 117. Fundamentals in Information Theory
  • 118. Some Fundamentals in Information Theory
  • 119. Fundamentals in Information Theory
  • 120. Some Fundamentals in Information Theory
  • 121. Fundamentals in Information Theory
  • 122. Fundamentals in Information Theory
  • 123. Fundamentals in Information Theory
  • 124. Some Fundamentals in Information Theory
  • 125. Fundamentals in Information Theory
  • 126. Some Fundamentals in Information Theory
  • 127. Fundamentals in Information Theory
  • 128. Fundamentals in Information Theory
  • 129. Fundamentals in Information Theory
  • 130. Some Fundamentals in Information Theory
  • 131. Classification and Regression Trees (CART)
  • 132. Goal
  • 133. Classification and Regression Trees (CART)
  • 134. Goal
  • 135. Classification and Regression Trees (CART)
  • 136. Some Fundamentals in Information Theory
  • 137. Fundamentals in Information Theory
  • 138. Some Fundamentals in Information Theory
  • 139. Classification and Regression Trees (CART)
  • 140. Goal
  • 141. Classification and Regression Trees (CART)
  • 142. Goal
  • 143. Classification and Regression Trees (CART)
  • 144. Some Fundamentals in Information Theory
  • 145. Fundamentals in Information Theory
  • 146. Some Fundamentals in Information Theory
  • 147. Classification and Regression Trees (CART)
  • 148. Goal
  • 149. Splitting Criteria for the Decision Tree
  • 150. Goal
  • 151. Splitting Criteria for the Decision Tree
  • 152. Goal
  • 153. Classification and Regression Trees (CART)
  • 154. Goal
0/154
Volume
1.0x
00:00/1:05:50