Current Proceedings on Technology
Yazarlar: Husniza Husni
Konular:-
Anahtar Kelimeler:Automatic speech transcription,Speech processing,Read speech,Automatic speech recognition
Özet: This paper proposes an automatic speech transcription for dyslexic children’s read speech using a speech recognition engine trained on lexical and language models specifically constructed based on their recorded readings. Automatic transcription of recorded speech is useful to facilitate speech processing for researchers. It could effectively increase efficiency of the process by eliminating manual transcription that is usually very time-consuming and subject to human errors. To automate the process, automatic speech recognition engine is required to perform the task of automatically labeling and transcribing speech into its corresponding phonemic and text representations, resulting in an individual phoneme and text files for each read speech transcribed. For this purpose, a total of 6112 speech samples were recorded from a number of ten dyslexic children reading aloud prompted isolated words in Malay. The speech samples are used to perform the automatic transcription and the accuracy is measured. Evaluation is conducted to determine whether or not the automatic transcription accuracy is acceptable for use in speech processing research. The findings reveal that automatic transcription is promising to be used for speech processing. It reduces time and effort as well as simplifies the manual phonetic labelling and transcription process.