Speech Recognition Model in Yoruba Language

Abstract

Speech has been widely known as the primary mode of communication among individuals and computers. The existence of technology brought about human-computer interfaces to allow human-computer interaction. While speech recognition systems have been developed rapidly over the last few years, many challenges have been encountered, such as language barrier, speech rate, environmental conditions, channel, and context of utterance. To enhance the power of speech recognition systems, it is required to design speech recognizers in local languages. There is a lot of development and research in the field of foreign languages but to enhance its power and utility for native people, it’s essential to use this technology in native languages. In this paper, we employ acoustic modeling and language modeling for the task of recognizing speech and also converting the speech to text by constructing sequential MFCC features from the frames of the audio sample. The result shows that concatenating Mel-Frequency Cepstral Coefficients (MFCC) features sequentially and applying acoustic modeling and language modeling provide a solution to the problem of recognizing speech and also converting the speech to text with an accuracy of 83%.