On-device implementation of an automatic Filipino speech recognition system

• Main Authors: Ang, Federico M., Guevara, Rowena Cristina L.
• Format: text
• Published: Animo Repository 2008
• Subjects: Automatic speech recognition; Speech-to-text systems; Systems and Communications
• Online Access: https://animorepository.dlsu.edu.ph/faculty_research/12343
• Institution: De La Salle University

This paper describes the preliminary results in an effort to develop a speaker-independent Filipino Speech-to-Text (STT) application on a smartphone. The system is composed of a front-end Symbian C++ application that was built to run on Symbian OS S60 3rd Edition smartphones. The application covers significant portions of a standard speech recognition system, namely, capture of speech input, feature extraction using PLP-RASTA directly converted to cepstral values, and posterior probabilities estimation using a multi-layer perceptron. Acoustic and language models were produced using different LM tools. To obtain the accuracy of the system, it was tested with a Linux-based decoder (OWAY) from SPRACHcore, a full source code release of speech recognition tools from ICSI, which was ported to run on a Windows-based desktop computer. Tests were done in a moderately noisy environment. System performance had average recognition rates of 53.94% and 6.15% for word-and sentence-level recognition, respectively.