Ibikunle Frank^1,, Katende James¹

Speech Recognition is the technology by which sounds, words or phrases spoken by humans are converted into electrical signals and these signals are transformed into coding patterns to which meanings are assigned. It has two main types: discrete word and continuous speech recognition systems. Each type can be further sub-divided into two categories as Speaker Dependent and Speaker Independent recognition systems. Speaker dependent system operates only on the speech of a particular speaker for which the system is trained, while the Speaker Independent systems can be operated on the speech of any speaker. The speech recognition system proposed here digitizes the isolated words spoken by a speaker and performs Mel Frequency ceptral analysis and other signal processing techniques on the digitized data. The processed speech signal is then passed on to a pattern recognition which takes action based on the type of command pattern received. Artificial Neural Network (ANN) is used as speech recognition engine. Two different corpora were collected of audio recordings of Yoruba, Igbo and Hausa language speakers, in which subjects read aloud different words. One of the collected corpora contained data with background noise and the other without background noise. The results obtained from simulation can be generalized to cater for larger vocabularies and for continuous speech recognition.

speech recognition, Artificial Neural Networks, MFCC, Yoruba, Igbo and Hausa languages