FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W940699370> ?p ?o ?g. }

• W940699370 abstract "The automatic speech recognition (ASR) field has become one of the leading speechtechnology areas nowadays. The research in ASR has always been emphasizing ondeveloping man-machine communication and promising in ease of use over thetraditional keyboard and mouse. The speech recognition task is simple to beidentified by human, but a very complex process for the machine to understand.Various methods have been introduced to develop an efficient ASR system. ANeural Network (NN) approach is one of the famous methods and widely used in thisfield. A Multilayer perceptron (MLP) is a popular NN model used in ASR field. Inthis study, a MLP with back propagation learning algorithm is implemented toperform the isolated digit speech recognition task for Malay language. However, oneof the current problems faced by MLP and most NN models in ASR field is the longlearning time. Besides that, the requirement to produce high recognition rate forisolated digit speech recognition system performed by MLP is also not trivialbecause it has been widely used in many applications. Thus, this study focuses onimproving the learning time and recognition rate of the MLP neural network for Malay isolated digit speech recognition system. This current study proposes threenew methods to fulfill the objective above. The improvement is made inpreprocessing and recognition phase. In preprocessing phase, a new endpointdetection method is proposed and it is known as variance method. This method isintroduced to overcome the disadvantages of the conventional method. Theobstacles in the conventional method are unstable and difficult to set the thresholdduring the silence detection. Hence, poor recognition rate is produced. Anothercontribution in the preprocessing phase is in normalization phase. Threenormalization methods are introduced to normalize the speech data beforepropagating to NN. The proposed methods consist of exponent, hybrid I and hybridII. These methods are compared with 4 widely used conventional normalizationmethods. These include range I, range II, simple and variance method. Theconventional methods have two limitations. The first is that some of the methods arevery slow in learning phase but produce good recognition rate such as variance andrange I methods. The second is that few of them are very fast in learning phase butproduce low recognition rate such as simple and range II methods. Therefore, thenew normalization methods are proposed to accelerate learning time and to producehigh recognition rate. In recognition phase, a simple novel approach is introduced toincrease the recognition rate. An adaptive sigmoid function is implemented toachieve this objective. A typical or fixed sigmoid function method is used inlearning phase. In the recognition phase, an adaptive sigmoid function is employed.In this sense, the slope of the activation function is adjusted to gain highestrecognition rate. This study emphasizes on 10 Malay words that comprise of “sifar”to “sembilan” (“0” to “9”). All utterances were recorded through single male speakerand each utterance was repeated 100 times. Thus the data set consist of 1000 utterances of Malay words. Four hundred data sets were split to utilize in the learningphase and the remaining 600 data for recognition phase. The TI46 standard data setwas used to evaluate the performance of the all proposed method and 10 Englishwords, consisting of “zero” to “nine” (“0” to “9”) are utilized throughout this study.Eight male and female speakers uttered each word 8 times. Hence, the total data setis 1600 for both speakers. The data set based on male and female speaker is trainedseparately. In this sense, four hundred male data sets were experimented duringlearning phase; meanwhile 400 data sets are kept as test data. The same approach isutilized in learning and recognition phase for female data sets. The Linear PredictiveCoding (LPC) is implemented as a feature extraction method to represent the speechdata. The experimental results show that the proposed endpoint detection (variancemethod) produced promising results in term of learning time and recognition rate.Meanwhile, the proposed normalization method has shown excellent results over allexperiments. The adaptive sigmoid function also successfully increased therecognition rate in the most of the experiments. Finally, from the overallexperiments, it can be concluded that the highest recognition rate for Malay data setis 99.83% with 82s convergence time. Meanwhile, for TI46 data set (female andmale data set), the yielded convergence time is 55s and 111s with the recognition rateof 96.75% and 94.75% respectively." @default.
• W940699370 created "2016-06-24" @default.
• W940699370 creator A5052771970 @default.
• W940699370 creator A5090108579 @default.
• W940699370 date "2004-08-01" @default.
• W940699370 modified "2023-09-27" @default.
• W940699370 title "Development of an Isolated Digit Speech Recognition Based on Multilayer Perceptron Model" @default.
• W940699370 hasPublicationYear "2004" @default.
• W940699370 type Work @default.
• W940699370 sameAs 940699370 @default.
• W940699370 citedByCount "0" @default.
• W940699370 crossrefType "dissertation" @default.
• W940699370 hasAuthorship W940699370A5052771970 @default.
• W940699370 hasAuthorship W940699370A5090108579 @default.
• W940699370 hasConcept C111919701 @default.
• W940699370 hasConcept C119857082 @default.
• W940699370 hasConcept C127413603 @default.
• W940699370 hasConcept C153180895 @default.
• W940699370 hasConcept C154945302 @default.
• W940699370 hasConcept C179717631 @default.
• W940699370 hasConcept C201995342 @default.
• W940699370 hasConcept C202444582 @default.
• W940699370 hasConcept C2780451532 @default.
• W940699370 hasConcept C28490314 @default.
• W940699370 hasConcept C33923547 @default.
• W940699370 hasConcept C34736171 @default.
• W940699370 hasConcept C41008148 @default.
• W940699370 hasConcept C50644808 @default.
• W940699370 hasConcept C9652623 @default.
• W940699370 hasConcept C98045186 @default.
• W940699370 hasConceptScore W940699370C111919701 @default.
• W940699370 hasConceptScore W940699370C119857082 @default.
• W940699370 hasConceptScore W940699370C127413603 @default.
• W940699370 hasConceptScore W940699370C153180895 @default.
• W940699370 hasConceptScore W940699370C154945302 @default.
• W940699370 hasConceptScore W940699370C179717631 @default.
• W940699370 hasConceptScore W940699370C201995342 @default.
• W940699370 hasConceptScore W940699370C202444582 @default.
• W940699370 hasConceptScore W940699370C2780451532 @default.
• W940699370 hasConceptScore W940699370C28490314 @default.
• W940699370 hasConceptScore W940699370C33923547 @default.
• W940699370 hasConceptScore W940699370C34736171 @default.
• W940699370 hasConceptScore W940699370C41008148 @default.
• W940699370 hasConceptScore W940699370C50644808 @default.
• W940699370 hasConceptScore W940699370C9652623 @default.
• W940699370 hasConceptScore W940699370C98045186 @default.
• W940699370 hasLocation W9406993701 @default.
• W940699370 hasOpenAccess W940699370 @default.
• W940699370 hasPrimaryLocation W9406993701 @default.
• W940699370 hasRelatedWork W115323076 @default.
• W940699370 hasRelatedWork W1520413003 @default.
• W940699370 hasRelatedWork W2104453471 @default.
• W940699370 hasRelatedWork W2151129782 @default.
• W940699370 hasRelatedWork W2154288735 @default.
• W940699370 hasRelatedWork W2160162224 @default.
• W940699370 hasRelatedWork W2160985694 @default.
• W940699370 hasRelatedWork W2312554498 @default.
• W940699370 hasRelatedWork W2315433741 @default.
• W940699370 hasRelatedWork W2352185638 @default.
• W940699370 hasRelatedWork W2352468006 @default.
• W940699370 hasRelatedWork W2377526959 @default.
• W940699370 hasRelatedWork W2574984907 @default.
• W940699370 hasRelatedWork W2587684252 @default.
• W940699370 hasRelatedWork W286004604 @default.
• W940699370 hasRelatedWork W2936078256 @default.
• W940699370 hasRelatedWork W35814158 @default.
• W940699370 hasRelatedWork W770188664 @default.
• W940699370 hasRelatedWork W2845591991 @default.
• W940699370 hasRelatedWork W2857054191 @default.
• W940699370 isParatext "false" @default.
• W940699370 isRetracted "false" @default.
• W940699370 magId "940699370" @default.
• W940699370 workType "dissertation" @default.