摘要 : Speech Signal Processing is one of the important branches in Signal Processing (SP). Speech Recognition (SR) is the process of conversion of speech signal-to-text conversion. It is carried out in two stages, training phase and tes... 展开 Speech Signal Processing is one of the important branches in Signal Processing (SP). Speech Recognition (SR) is the process of conversion of speech signal-to-text conversion. It is carried out in two stages, training phase and testing phase. This paper presents Implementation and verification of speech recognition algorithms for isolated words has been carried out using Mel Frequency Cepstral Coefficients (MFCC) as a feature extraction technique. Using this algorithm, the cepstral coefficients are calculated on Mel frequency scale then Vector Quantization (VQ) carried out which is used as a speech modeling technique. VQ is used to prepare a single template/codebook of each utterance called as word model from given set of training data samples of the given utterances (100 samples per utterance) during training phase. In the feature matching, or template matching, similarity measures are calculated by Euclidean Distance(ED) criterion. Because of high accuracy of used algorithms, the accuracy of this voice command system is quite high. MATLAB 7.1 is used for implementation of the algorithms. Sound Forge 9.0 & Gold Wave is used as speech analysis Software. Speech recognition rate using the above mentioned algorithms tested on 'TI-46' database and locally prepared 'DYP' database are 96% and 92% respectively. Results are evaluated and compared by MFCC and VQ algorithm by additive Gaussian noise of 15dB, 10dB and 5dB level in the signal. 收起