Grammar guided automatic speech system for Malay language / Radde Idzwan Abdul Razak
Speech recognition is difficult because of the analog nature of the speech itself. In speech recognition, a system is trained to receive human speech, recognize each word in the speech and transform it into text. There are many existing Automatic Speech Recognition (ASR) system that able to recogniz...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2009
|
| Subjects: | |
| Online Access: | https://ir.uitm.edu.my/id/eprint/98213/1/98213.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Speech recognition is difficult because of the analog nature of the speech itself. In speech recognition, a system is trained to receive human speech, recognize each word in the speech and transform it into text. There are many existing Automatic Speech Recognition (ASR) system that able to recognize English language speech with high accuracy. Unfortunately, an ASR system normally focuses on only one language. Today, there are no or lack of Malay language recognizable ASR system. Thus, an ASR system that is capable of recognizing Malay words would be beneficial especially in Malaysia. In order for the ASR system to transform speech signal into a specific word, the system requires a database which store list of words with each word pronunciation. This database is called a dictionary. A dictionary is use by an ASR system as references to find words match to the received speech signal. In order to be able to recognize spoken Malay words, a dictionary must contain a list of Malay words and each word must be associated with its pronunciation. Thus, this research is focusing on building Malay dictionary for an ASR system. The dictionary built in this research is meant to be used for ASR system based on Sphinx-4 framework. Each word is tested for recognition accuracy on the ASR system prototype. Trial and error method was used to produce word’s phonemes that are most accurate. Well defined words in the dictionary do not guarantee high recognition accuracy. Word Error Rate (WER) typically increases when the size of dictionary is increase. Grammar implementation would reduce the number of words to be combined to construct a sentence. Thus, it increases the chances of accurate recognition with large dictionary for continuous speech. The finding of this research shows that Sphinx-4 is a good framework to be use with Malay language ASR. The outcome of this research proved that grammar able to increase recognition accuracy of Malay language ASR. |
|---|