00000ctm a22000004a 4500 UP-99796217603116358 Buklod 20230503092542.0 m |o d | ta 070718s2000 xx r |||| u|eng d (iLib)UPD-00000179802 DENGII LG 993.5 2000 C65 B45 Baquiran, Eric P. The Applicability of linear predictive coding and vector quantization to Filipino speech recognition Eric P. Baquiran ... [et al.]. 2000. vi, 41 leaves col. ill. Thesis (B.S. Computer Science)--University of the Philippines Diliman. This thesis examines how Linear Predictive Coding and Vector Quantization can benefit a small vocabulary, speaker independent, isolated Filipino speech recognition system. It uses a template-based approach, in which an unknown word sample is compared against a number of stored word templates to find the best match. A speaker independent isolated speech database of 10 words was created using a training set consisting of 160 training words (16 utterances per digit sample). Voices of Filipino males between the ages of 15 to 25 were recorded in low-noise environments. The samples were preprocessed using FIR filtering. Excess leading and trailing silences were then removed. Finally, each sample was linearly time warped. Individual code-books for each digit were created using the K-means clustering algorithm applied to the spectralcoefficients derived from each training word using LPC Analysis. The codebooks were then concatenated to form the global codebook. Preliminary testing showed that this approach performed with good recognition accuracy (over 85%). However, when the approach was extended to a vocabulary of 101 words (the English words for the numbers 0 to 100), accuracy fell greatly (below 15%). It is recommended that improvements be done to the quality of the training set and the preprocessing steps (endpoint detection, time warping) to increase the accuracy when applied to larger vocabularies. Automatic speech recognition. Speech processing systems. Speech processing Digital techniques. Speech synthesis. Filipino speech recognition. Linear Predictive Coding and Vector Quantization. Speech recognition system. Belleza, Liberty Irene A. Catoy, Marvin N. Dideles, Myra Colina B. Thesis FI UP UPD DENG-II LG 993.5 2000 C65 B45 Thesis